A Review of the High-Mix, Low-Volume Manufacturing Industry

Abstract

The high-mix, low-volume (HMLV) industry has seen growth in the need for product customisation with research to increase manufacturers’ flexibility for the variation in market demands. This paper reviews 152 documents from 2000 to October 2022, discussing work related to HMLV production. From an industrial perspective, this paper analyses the industries with HMLV, revealing production sectors and research areas, categorising the developed work, type of validation, and applications. The results show that most work is not industry-specific, with production planning as the central aspect of the research. While other parts of the production processes and value chain received less attention, the semiconductor and electronics industries are the two most researched with substantial validation, leaving gaps in other industries. Earlier work primarily focused on the theoretical development of production planning; however, the development of Industry 4.0 technologies advocates decision support systems for reactive production planning. This period sees the rise in robotics and automation, with improved robotics capability and human—robot collaboration. Assembly assistance systems were developed for manual production to aid operators in managing the variety of information. This paper serves as a reference for the HMLV manufacturing industry in a structured manner while identifying potential for future research in this field.

1. Introduction

Changing market trend has created demands for product differentiation and personalisation. This trend shifts manufacturing from high-volume, low-mix production to high-mix, low-volume (HMLV) production [1,2]. In a high-mix production, the non-repetitive products were made-to-order with distinct production routes on the production floor, each requiring its own setup and processing time in the factory [3,4,5,6].

HMLV has a high degree of customisation [3,7] that inevitably creates non-standard product routings on the shop floor. Therefore, accommodating the variability is critical for manufacturers in the HMLV industry. The HMLV industry includes part of the make-to-order (MTO) manufacturers that are in the versatile manufacturing companies (VMCs) category [8]. In the VMCs’ HMLV environment, each order is completed individually, with a high variety of products and variable demands, which themselves are manufactured in small batches with little repetition. Unlike repeat business customizers (RBCs) of the MTO, which produces customized products continuously over the contract period, VMCs cannot entice customers to build a more stable, predictable and committed relationship [7].

1.1. Industry 4.0 for Production Flexibility

Industry 4.0 is a concept of an interconnected system of real and virtual factories represented by cyber-physical systems (CPSs). A cyber-physical system is a system of collaborating computational elements that control physical entities. The CPSs are physical and engineering systems whose operations are monitored, coordinated, controlled, and integrated, based on the computing and communication core. Embedded system relies on sensors to record data and manipulate physical processes with the help of actuators operated in digital networks. With the core elements of connectedness, smart machines and products, decentralisation, big data, and cybersecurity [9]. These elements deliver intelligence, connectedness, and responsiveness to changes [10].

This flexibility of the Industry 4.0 concept also provides vital capabilities for HMLV manufacturers to handle the operation on small batches, down to batch-size-one. Industry 4.0 envisioned real-time, zero-setup-time production flexibility to meet the demand for personalisation and mass customisation [11]. Two of the pillars of Industry 4.0 best reflect the production flexibility in HMLV—additive manufacturing and product identification, and traceability.

One of the vital Industry 4.0 technologies for flexible production is additive manufacturing and 3D printing technology [12]. Three dimensional design data enables components to be built in layers by depositing material in fine powder form. Despite the initial high investment, additive manufacturing has a cost advantage for small volumes, with the capability for complex design, shorter time-to-market, and flexibility down to batch-size-one.

Traceability and product identification are the fundamentals of Industry 4.0 production flexibility. The ability to assign a unique ID to each component enables real-time control of the value chain over the product life cycle. Unique IDs make individual components identifiable throughout the production process; this allows dynamic, efficient production path planning down to individual components. Information on each component, such as origin, storage, state, and location, is retrievable instantly.

With the increase in costs in a high-wage economy, automation and robotic production has become more attractive for HMLV’s flexible manufacturing system [13]. Flexible end effectors, advanced integrated peripheries, and easy programming are key enablers that make collaborative robots flexibly attractive for the HMLV industry.

1.2. Shop Configuration of HMLV

Most HMLV production is in a job shop environment where the routing sequences are random to allow the flexibility of the job to start and finish at any work centre for customisation [7]. Shop configurations were unlikely to lie at the extremes; instead of a pure job shop environment, a dominant flow direction may exist [14]. This form of job shop, which allows multi-directional routing with a dominant flow direction, is called a generalised job shop.

1.3. Challenges in HMLV

As consumers increasingly demand customised tailored products, manufacturers leverage the trend to differentiate themselves from their competitors. However, the HMLV manufacturing environment differs from conventional mass production’s high-volume, low-mix environment. Such an environment shows the following characteristics, which are challenging to the operational team: [15]

Many different product numbers are produced in small quantities;

Various routing for all of the products [16,17];

Job shop manufacturing environments [18,19];

High variance in cycle time depending on the product type;

High variation of demand in different parts.

1.4. Reviews in HMLV Topics

HMLV topic is often discussed from the perspective of production planning. However, Industry 4.0 technologies, such as robotics, intenet of things (IoT), and mixed reality, have renewed the discussion of new possibilities for the HMLV.

This paper aims to review HMLV production from the manufacturing industry perspective; to investigate the industrial sectors and production areas involved in the HMLV type of production and operation. The authors hope to uncover research that addresses HMLV production issues, categorising them according to their field of improvement, and thus exploring the future research gap.

A recent systematic literature review on the HMLV industry was published by Tomašević, Stojanović, Slović, Simeunović and Jovanović [4] on lean in HMLV. On the topic of lean, Miqueo, et al. [20] also completed a systematic review of Lean Manual Assembly 4.0 to investigate the improvements of Industry 4.0 technologies in manual assembly operations. An industrial-specific survey on HMLV was carried out in the defence and aerospace industries [21]. Meanwhile, HMLV-related technologies covered by some systematic literature reviews were technology-specific, such as robotics and automation [22,23] and augmented reality [24,25].

Therefore, this paper covers all aspects of the HMLV industry, including their manufacturing design and planning, management, and different types of production—robotics, automated, and manual production. This paper analyses the manufacturing industry involved with HMLV production.

2. Methodology

This section describes the methodology for systematic literature review. It follows the method based on Denyer and Tranfield [26] ‘Producing a systematic review’. It follows a six-step methodology.

2.1. Step 1: Planning

Systematic literature review starts with formulating research questions to determine the broad topic areas to be investigated. These are further narrowed down to specific areas for in-depth analysis. For this, the research questions are as follows:

• Q1: What has been the trend of HMLV research for the past two decades?
• Q2: What are the current areas of interest in HMLV research?
• Q3: How are digital tools being used in HMLV for production management?
• Q4: What is the extent of simulation being tested on the industry?
• Q5: How has research in automation and robotic production contributed to HMLV?
• Q6: How is manual production optimised for HMLV?
• Q7: What is the future direction for HMLV research?

The objective is to discover the underexplored gap in the HMLV research for future research.

The databases selected for the search are as follows:

Scopus;

Web of Science (WoS).

These two databases are selected for their broad coverage of the indexing of research materials and documents across different full-text databases—to minimise the exclusion of any potential materials from various publishers. The reference manager used was Endnote; Microsoft Excel was used for data extraction and analysis, and VOSviewer was used for network mapping.

2.2. Step 2: Search

The search strings for systematic searches were decided after referring to other literature review papers on HMLV and manufacturing technology [4,25]. The search filter included technical and review journal articles and conferences that were published between 2000 to October 2022. The search string is:

(HMLV OR (High Mix Low Volume) OR (High-mix Low-Volume))

AND

(Manufact∗ OR Production OR Industr∗ OR (Industrial Application∗) OR logistic∗ OR maintenance OR training OR quality OR shopfloor OR warehouse OR assembly)

As the keyword ‘high-mix and low-volume’ is written differently across authors, the search string included different ways the HMLV keyword might appear in the literature. The ‘AND’ association rule combines the first part of the string with the second part. The second part limits the results to research in manufacturing and production systems, as HMLV could be researched in other fields, such as medicine and marketing.

The range of the literature review is from the year 2000 to October 2022 due to a lower number of publications befor000 and becoming less relevant as technology progresses.

Table 1. Inclusion and exclusion criteria of the database results.

Inclusion Criteria	Exclusion Criteria
An article is a journal or conference paper with relevance to the topic	Not in English
Published between 2000 to October 2022	Documents not accessible from online databases
	Not manufacturing or engineering related

shows the criteria being considered in the database results filtering.

2.2.1. Search String

A list of documents was obtained from the two databases with search strings without reading the title and the abstract. As potential articles might be missed from database searches [27], other relevant literature was added manually. Manually added papers are subjected to the same treatment as those from databases in the following steps.

SCOPUS:

(TITLE-ABS-KEY ((hmlv OR (high AND mix AND low AND volume) OR (high-mix AND low-volume))) AND TITLE-ABS-KEY ((manufact* OR production OR industr* OR (industrial AND application*) OR logistic* OR maintenance OR training OR quality OR shopfloor OR warehouse OR assembly))) AND PUBYEAR > 1999 AND (LIMIT-TO (DOCTYPE, “ar”) OR LIMIT-TO (DOCTYPE, “cp”)) AND (LIMIT-TO (SUBJAREA, “ENGI”) OR LIMIT-TO (SUBJAREA, “COMP”) OR LIMIT-TO (SUBJAREA, “ENVI”) OR LIMIT-TO (SUBJAREA, “BUSI”) OR LIMIT-TO (SUBJAREA, “ENER”) OR LIMIT-TO (SUBJAREA, “MULT”))

Web of Science:

((TS = (HMLV OR (High-mix Low-Volume) OR (High-mix/low-volume)) AND TS = (Manufact* OR Production OR Industr* OR (Industrial Application*) OR logistic* OR maintenance OR training OR quality OR shopfloor OR warehouse OR assembly)))

Timespan: 1 January 2000 to 31 October 2022 (Index Date)

2.2.2. Co-Citation Analysis

A bibliometric method is used to eliminate irrelevant papers, which identifies the overall knowledge structure of the search result from databases. Co-citation analyses the number of times two journals were cited together, revealing their relationship. This method identifies the ‘structural knowledge group’ in HMLV research [28].

VOWviewer is used to generate a bibliometric map of the co-citation of the search result. As the WoS database is more selective and covers only peer-reviewed publications, only its search result was used by VOSviewer to explore the relevant research topics for result screening.

The co-citation network map generated in Figure 1 shows four significant clusters of publication journals roughly corresponding to the area of the research topic. It includes (i) computing and industrial engineering (red), (ii) advanced manufacturing and robotics (green), (iii) production and operation management (blue) and (iv) automation, semiconductor, and simulation research (light green).

2.2.3. Search Results

From the research clusters, irrelevant papers were removed after going through the title of each publication. Useful documents that were known to the authors were manually added to the screening process for the following step.

Table 2. Results from the searches.

		No of Documents
Scopus	Paper found	661
	Irrelevant papers	397
Web of Science	Paper found	182
	Irrelevant papers (after title and abstract screening)	66
Duplicates		32
Documents Obtained		355
Manually Added		6

shows the results of Step 2.

2.3. Step 3: Title and Abstract Screening

The title and abstract of each document were screened to filter out excluded documents and duplicated documents, based on the inclusion and exclusion criteria in Table 1, as the results from the databases showed that not all irrelevant documents were excluded. Additional documents, which are known to the authors, related to the topic were also added manually.

To further refine the criteria in this process, additional criteria include:

Exclusion Criteria:

Duplicated documents;

The relation to HMLV production is unclear.

Inclusion Criteria:

Studies related to MTO with HMLV.

From the 355 previously obtained documents, 225 documents were chosen after the title and abstract screening.

2.4. Step 4: Introduction and Conclusion Screening

Each document’s introduction and conclusion were screened for their relevance according to the inclusion and exclusion criteria. This step narrows down the documents with a thorough selection process; 128 documents are obtained after this step.

2.5. Step 5: Evaluation

In addition to the inclusion and exclusion criteria, the quality of the publications was evaluated. There were three criteria for the articles to be selected:

Methodology elaborated;

Results presented;

The article is relevant to the research questions.

The documents that did not satisfy the criteria may still be selected if the paper presents interesting ideas and frameworks that might not be executed. Following the three criteria above, Figure 2 shows how the final 152 documents are selected for data extraction.

2.6. Step 6: Extraction

For this systematic information extraction, a table is built.

Table 3. Excerpt of the data extraction of each document in step 6.

Authors	Title	Sector - Area	Aspect - Aim	Methodology	Variables	Significance	Limit/Gap	Tested on Industry
Abu-Samah, A., et al. (2017)	Bayesian-based methodology for the extraction and validation of time-bound failure signatures for online failure prediction	General - Maintenance	Failure Prediction - Max Resources Utilisation	-Math Modelling -Simulation	- failure probabilities - Predictability index	- Bayesian Network for failure estimation - User exp input for identification of failure predictors, patterns, and warning-time - Validation at one module of the SI industry	- Multiple failure detection at the same interval is not supported	Semiconductor
Bohnen, F., et al. (2013)	Systematic procedure for levelling of low volume and high mix production	General - PO Management	P. Scheduling - Max Resources Utilisation	-Math Modelling -Simulation	-EFEI (Every family every interval) - Capacity slot -overall capacity for changeover of levelling families	- Systematic procedure for HMLV levelling based on Group Tech - Families are used for levelling - Dev software toolkit.	- Simulation only - No numerical comparison with other levelling methods.	N/A
Hibino, H., et al.	A study on lot-size dependence of the energy consumption per unit of production throughput concerning variable lot-size	General -PO Management	P. Scheduling -Minimise Lead Times -Minimise makespan -Max Resources Utilisation	-Simulation -Math Modelling	- Energy consumption per unit throughput - Lot Size ratio - Average lot size -Setup time	- Production management that considers lot size for energy consumption/unit production throughput for variable lot sizes	- Simulation on semiconductor industry data. - Application of formulation to lot-size optimization	Semiconductor (dataset)

shows an extracted representation of the table. The type of data extracted was based on previous literature reviews, to arrange the data systematically.

2.6.1. Sector of Research

It is the sector of industry of the research, as the particular research targeted improvement of HMLV manufacturing processes of a specific industry, i.e., electronics manufacturing. Otherwise, research that aims at the non-specific HMLV industry was categorised as “General”.

2.6.2. Industrial Area of Research

It categorises the area of research within the manufacturing system. It includes production and operation management, production, quality, maintenance, logistics, training, ICT, robotic production, sustainability, and product development.

2.6.3. Aim of the Research

Studies that aim to improve the manufacturing processes were grouped under production technologies. In contrast, research that optimises the production system with a multi-aspect approach is grouped under the production planning category. Other specific research was categorised accordingly, i.e., order fulfilment or failure prediction.

2.6.4. Methodology

The methodologies used to achieve the research aim were categorised in this column. It includes mathematical modelling, simulation, case study, user testing, and framework.

The classification of user testing includes laboratory experiments and actual field experiments. The distinction between them for this literature review is that field experiments were in the factory environment. Meanwhile, user testing was carried out on related factory personnel, such as operators, technicians, and supervisors. If there were field experiments, the sector where they were conducted would be separately noted in another column to investigate the extent of their industry application.

2.6.5. Variables

It describes some of the critical measurements being considered in the research. Measurements include cycle time, throughput, lot sizes, and task performance.

2.6.6. Challenges and Limitations

These data are collected to address research question 7 (Q7: What is the future direction for HMLV research?). Therefore, all of the limitations and challenges noted by the authors are recorded in this column.

2.7. Research Keyword Co-Occurrence Analysis

In a keyword co-occurrence network, nodes are the keywords, while the edges show the relationships among nodes. Two keywords are counted as co-occurring when they appear on the same title or abstract. The closer the distance between two nodes, the higher the rate of co-occurrence [29].

Following the analysis of the keyword co-occurrence network, the knowledge cluster and the hotspot of research areas can be identified from the final extracted publications. This is shown by the size of nodes and their degree of centrality [30]. Table 3 shows the results of Step 6.

3. Current Status of HMLV Research

3.1. Number of Publications over the Two Decades

Figure 3 illustrates the number of documents obtained for the title and abstract screening (unfiltered documents) versus the final publications used for the data extraction (filtered documents). Over the selected period, around 40% were used for data extraction.

The distribution of the extracted publications also shows three distinct phases: 2000–2008, 2009–2016, and 2017–October 2022. Therefore, the keyword co-occurrence analysis was carried out to identify the topics of interest during these phases.

3.1.1. Publication Keywords Co-Occurrence

2000–2008 Phase

The keywords are grouped into two clusters in the earliest time segment of the literature search in Figure 4. The red cluster denotes work related to printed circuit board assembly (PCBA), while the green cluster relates to manufacturing simulation. The research keywords showed a considerable emphasis on the research on manufacturing printed circuit boards. Algorithms and simulations work was for optimisation in the production and operation management challenges, such as scheduling and levelling.

2009–2016 Phase

From 2009 to 2016, Figure 5 shows that new research topics emerged as there was more interest in automation and robotics in manufacturing industries, as the concept of Industry 4.0 could bring more flexibility for HMLV manufacturing.

Earlier research, in the 2010s, mainly focused on algorithms, such as genetic algorithms, to solve production and operation management problems. However, after 2013, the implementation of HMLV in the automated production system became the focus. For example, the topic of production control involving automation and robotics was popular. Maintenance and product design were the two new important aspects of the research in an HMLV environment.

The research focuses also moved from PCBA to the challenges of the broader semiconductor industry. The HMLV job shop nature of the industry requires production planning and control. Two main topics were frequently discussed—maintenance and automation.

2017–2022 Phase

Figure 6 shows the topics of interest from 2017 to October 2022 with five main clusters. Yellow clusters are mainly on robotics and automation. Manual assembly and lean topics are shown in red. The dark blue cluster focuses mostly on production planning and control methods, such as order release and workload control with simulations. At the same time, light blue shows the manufacturing consideration of production resource management, such as scheduling, costs, decision making, and assembly machines. The green cluster shows Industry 4.0-related topics and the semiconductor industry. Meanwhile, purple relates operation research and management science with algorithms and optimisation.

The development in 2017–2022 in Figure 7, saw a growing interest in research in the popular topics of Industry 4.0—robotics, artificial intelligence, the internet of things, and additive manufacturing.

Research for industrial robotics and automation for flexible manufacturing were made to make robotic programming easier and more flexible. Techniques, such as deep learning, is a popular keyword in these studies. The concept of Industry 4.0 also hoped to improve the productivity of manual and semi-automated manufacturing. Therefore, human–robot collaboration was an area of interest to cater for the flexible HMLV manufacturing environment.

Instead of algorithms, the researchers looked at a broader ecosystem of production planning. Collecting production data in real-time to build big data, enables a more comprehensive data analysis for decision-making systems. Together with forecasting and reinforced learning, these were the three keywords that came up after 2020.

3.2. Industrial Sector of Research Publications

From Figure 8, seventy-one percent of the documents obtained address the HMLV industry in general. This includes production planning, scheduling, sequencing, flow management, and layout design.

The semiconductor industry (12%) gathered significant research interest due to the nature of wafer manufacturing with a high variation of products. Next, the electronics manufacturing industry (9%) is also extensively researched, where most of them are concerned with optimising production for PCBA.

The engineering support industry, such as contract manufacturers for machined parts and the aerospace industry, are also involved in HMLV manufacturing. Other industries, such as automotive, furniture, apparel, food, and construction, were topics in one research article each.

3.3. Research Area in the Manufacturing System

Identifying the research areas and the industry involved in the literature showed that most HMLV research problems were generalised non-industry-specific topics. This is denoted as “General” in Figure 9. At the same time, common industry-specific topics were identified independently.

The data show that the two most prominent industries with the most interest vested are the semiconductor industry and the electronics industry. Production and operation management research (‘PO management’) were the dominant researched topic, while less research was carried out on other parts of the production system.

3.4. Common Methodologies Used

Shown in Figure 10, the most common methodologies used in the research for HMLV improvements are simulation, mathematical modelling, and an algorithm-based solution for production planning. With a considerable rise over the years, 114 out of 118 documents found had used this approach.

While earlier research trends were in simulation and algorithm-based research to solve HMLV problems, there is a growing trend after 2013 for actual user testing of the research. These tests include real-world applications of their research into their selected manufacturing partners. The areas of this industrial testing will the elaborated on in Section 5.4.3.

4. Optimisation of the HMLV Layout and Flow

4.1. Production Line Design

HMLV manufacturing has a variety of raw materials, processing times, procedures, and flow, which require flexible assembly lines for product mix and absorb volume fluctuations.

As such, the use of mixed-model assembly lines (MMALs) and mixed-product assembly lines (MPALs) were featured in a case study by Asadi, et al. [31] in a heavy machinery manufacturing industry. It pinpointed the low level of product modularity as the main hurdle in MPALs. Therefore, establishing a common assembly sequence and defining similar module contents across product families would be the goal; to transition from semi-automatic MMALs to MPALs with a flexible assembly system for higher mix flexibility.

A hybrid production system that enables multiple product flows was also proposed [32]. The authors proposed an express line for high-running products alongside a regular production stream for implementing hybrid lead times in production lines [32]. In the apparel manufacturing industry, Moin, et al. [33] proposed a modular production system with a progressive bundling system (PBS) that reduced costs while coping with volatile market behaviour.

4.2. Layout and Flow Optimisation

4.2.1. Cellular Manufacturing

The layout design focuses on optimising product distribution and machine arrangement with mathematical modelling and simulation [34]. Group technology (GT) is a method where components are identified according to size, shape, and the type of processing required. With that, it optimises similarities in product design and manufacturing processes, allowing greater product variety at low volume. Cellular manufacturing (CM) applies the concept of group technology by assigning a similar group of parts into machines/manufacturing process clusters called cells. This shortens the setup times, in-process inventory, tooling, and improves product quality. Group technology approach was used to incorporate new parts and machines into the existing CM setup

Bhandwale and Kesavadas [35] devised a group technology model to incorporate new parts and machines into the existing cellular manufacturing system. The model uses a modified Boolean matrix concept to consider two cases. First is the inclusion of a new part family with all its operations on existing machines, where no new machines are added. This is where a company introduces a new variant of an existing design. The second case accounts for introducing a new part family with some operations on machines not in the existing layout; as such, new machines need to be introduced. This would apply to the company that introduces a new technology in the design.

Smed, et al. [36] developed mathematical formulae to solve group setups with minimal feeder change (GSMFC) problems. The devised model would have to optimise between two competing objectives: (1) to keep the number of product groups low and (2) to minimise the number of required feeder changeovers. The formulae included weightage, which has a natural interpretation of real-world situations, and different scenarios were analysed with changes in the weightage. From there, heuristic algorithms were devised to solve the GSMFC problem efficiently.

Simulation is also carried out on the effect of HMLV manufacturing with cellular manufacturing (CM) in the presence of demand fluctuations. Julie Yazici [37] has shown that cellular manufacturing provides faster delivery than job shop when the volume flexibility increases. CM also increases routing flexibility; the simulation model shows that the high-volume flexibility is balanced with high routing flexibility. By accounting for volume, routing, and labour flexibilities, the simulation shows CM’s benefits in reacting to the volatile demand in HMLV.

The implementation of CM has also been described by Karaulova and Shevtshenko [38]. It provided a framework and created a DES that defined the optimal time distribution for universal work cells with data analysis and a similarity analysis, followed by a work-cell definition for every product type and balancing of the workstation. This serves as an analysis tool for suggesting new factory layouts and redesigning the replenishment processes as a practical guide for work-cell implementation for enterprises.

In the context of Industry 4.0, Muniraj, et al. [39] presented a novel model of digital twins for manufacturing cells and a graph node network (GNN) for production routing. With a robotic operating system (ROS 2), individual digital twins are represented as nodes to create a decentralised and distributed architecture. With the GNN, the most time-effective path to the destination is calculated with the BFS algorithm. Meanwhile, Dijkstra’s algorithm is used to determine the shortest path when the paths are weighted.

4.2.2. Flexible Manufacturing Systems (FMS)

The flexible manufacturing system (FMS) is one of the concepts which manufactures a variety of products from a selected range of parts without batch production or time lost in reprogramming [40], while minimising reconfiguration time. The typical row layout with a straight-line flow is suitable for an automated assembly system with a well-defined work sequence for all work units. However, the length of the line may limit the movement of operators and materials if it is too long. The U-shaped line was developed to solve this by allowing the supply of material from outside to the operator inside. It also enables multi-functional workers to be mobile, performing tasks on both sides of the U-shaped layout.

Another approach is the loop layout, which is based on the straight-line flow that allows to serve as loading and unloading location at the same place: this is useful for pick-and-place on the material handling system. However, in the HMLV industry, there is no fixed material flow pattern. An open field layout is therefore presented as the random order form of FMS, which allows flexibility in the process sequence. This layout requires a greater number of machines and material handling. Therefore, an open field circular FMS layout is being proposed by Alduaij and Hassan [40], with the integration of robot-centred cells for more flexibility while being cost-effective at the same time. The face centred composite design (FCCD) approach is used to generate 40 scenarios that are used to numerically investigate the effects of layout, production quantity, variety, and complexity with minimal production costs. Then, linear programming (LP) is used to run optimisation models with layout-dependent machine distance algorithms individually. From there, a regression model is developed to provide quick estimates that negate the need to run intensive computational numerical models.

Meanwhile, Ojstersek et al. (2020) investigated the sustainability of the FMS. A four-level architectural model was used to describe the optimisation problem of HMLV production, using sustainability parameters in their mathematical modelling. The model optimised energy consumption and scrap percentage with a simulation study of a flexible job shop scheduling problem (FJSSP).

4.2.3. Flow Management

In HMLV environments, two critical aspects must be considered in implementing the flow in the manufacturing system. First, the variety and volume of products mean that production lines and cell construction differ from low-mix high-volume environments. Equipment and cells are shared among various tasks, in contrast with highly dedicated lines and cells. Next, balancing and stable line production are more complicated when subjected to customer diversity and demand uncertainties.

In static flow paths, pull production policies can be established, such as Kanban or constant work-in-process (CONWIP). Guan, Peng, Ma, Zhang and Li [16] suggested that the theories of constraints (TOC) and drum-buffer-rope (DBR) TOC/DBR can be applied to HMLV’s dynamic flow path. The control mechanism of TOC/DBR mainly consists of bottleneck detection, production planning on the bottleneck resource, setting and management of the buffer, and control of the production task time. This devised method hopes to better cope with the capacity imbalance and avoids cycle time and WIP increase caused by frequent flow interruption, due to any traditional pull production policy in HMLV environments.

Peng et al. (2008) developed a mathematical model to optimise adaptive flow paths according to product families to account for flow path management in a dynamic formation corresponding to different product families. Their mathematical model addresses the lack of consideration in machine sharing in the traditional approaches for cell design for part families. Therefore, the model considers variables that define the conditions for machine sharing.

Flow management in the semiconductor industry was a concern due to the variable nature of the wafer testing environment. Traditionally, the established layout baseline in semiconductor production is in the job shop manufacturing style; However, with customers demanding more integrated functions on a single chip, coupled with high innovation speed and short product life cycles, Keil, et al. [41] leverage the advantages of flow production. The developed virtual time-based flow principle (VTBFP) had the idea of using the time aspect of the flow production to logically overlay the job shop with VTBFP. This meant matching the flow times and capacity balancing of consecutive process steps for the continuous flow of material.

Lange, et al. [42] continued the work on the VTBFP approach by applying DES to design the production control for flow with the use case carried out on the wafer test facility. Unlike flow production, the wafer test has varying process times, for which a pure synchronised method is less appropriate. The VTBFP simulation has shown promising results in a partly VTBFP clocking setup. Setups time were reduced, and the optimum window was improved from 15–30% to 30–55%.

4.3. Loading Jobs/Job Release

Due to the dynamic production flow, the HMLV industry often has a job-shop environment. CONWIP and paired-cell overlapping loops of cards with authorization (POLCA) are two of the most common job release techniques for job shop with complex material flow discussed in the literature. These job release systems control the amount of work in a work centre, control the lead time, and monitor the backlog.

Constant work-in-process (CONWIP) is a continuous shop floor release method. The workflow is regulated by job number-specific cards, rather than part-number-specific cards. Therefore, the card stays with the specific product or batch throughout the process, providing a way to handle high variations. However, the CONWIP hierarchical control architecture may need items to be grouped into families of common routings, which is limited in HMLV.

To set the CONWIP limit in HMLV, Khan and Standridge [43] devised aggregate modelling to reduce the development time and computational requirements of HMLV simulations. The VSM approach is used to create an aggregate simulation model, which is used to set the limit of the CONWIP dynamically.

Eng and Sin [44] developed a DES for evaluating single-loop and multi-loop CONWIP control mechanisms. With a simulation based on the end-of-line process of a semiconductor company, a multi-loop approach shows better robustness with a lower cycle time and easier implementation. With the multi-loop system, cards could be returned faster, and starvations caused by bottleneck loops could be minimised.

In a job shop environment, CONWIP may lack the workload balancing required [18]. Frazee and Standridge (2016) evaluated this, which showed that CONWIP controls the maximum throughput at a lower WIP level than POLCA. Both CONWIP and POLCA in the simulation have an equivalent lead time. However, POLCA prevents WIP accumulates in one area, which CONWIP was unable to accomplish.

POLCA is a push-pull card-based signalling system that aims to reduce lead times, cut production costs, increase delivery date adherence, and reduce scrap and rework. This system was mainly devised for highly engineered production, small batches and high product variety [45]. In POLCA, cards are in pairs where one card travels in the direction of the job, and the other returns the information. This approach might be useful in a job shop multi-directional flow and variable routing by placing two cars between a pair of cells to allow work and feedback data to travel in both directions [7].

Variations of POLCA with a centralised and decentralised job release were also compared [46], showing centralised job release benefits in general flow shop and pure flow shop environments.

Workload control (WLC) is a production planning and control mechanism devised for a high variety of production, such as the MTO industry. By using a pre-shop pooling of orders to reduce shop floor congestion, shop more is made more manageable with a series of short queues. The shop’s performance is stabilised by making them independent of variations coming in from the order stream.

In most WLC cases, the jobs are only released onto the shop floor if the released workload levels do not exceed pre-set maximum limits, ensuring jobs do not stay in the pool too long to reduce lead times and meet delivery date objectives.

The WLC is based on two control mechanisms: input control (IC) and output control (OC). IC uses information about jobs in the system to decide when incoming orders can be released to the production process—order review and release (ORR). At the same time, OC uses information about the production capacity to regulate the workload in the system.

Costa, et al. [47] worked on integrating the order review and release (ORR) and output control (OC) with information about production capacity. Considering worker allocation as part of the OC mechanism, this work considered different labour-related parameters, such as flexibility and efficiency levels for workers. It reduces worker idling and the number of workers required while minimising the number of worker transfers.

As one of the challenges for manufacturers in HMLV is to provide the necessary master data quality for the input in ERP/MES system, this limits the system’s capability to plan and control tasks efficiently. Messner, et al. [48] proposed a closed loop between the shop floor and ERP for a better system integration with the centurio.work program. This closed loop is realised by connecting different systems, machines, and employees to gather contextualised data.

4.4. Value Stream Mapping (VSM)

The value-stream approach investigates steps where value is added and not added in the entire production process, including the supply chain. This lean flow approach using VSM is difficult to be adapted in HMLV due to the complexity of identifying and mapping value streams in a manufacturing environment with complex product structure and routing.

Thomassen, Alfnes and Gran [5] combined classic VSM methods with lean methods for engineer-to-order (ETO) production systems with HMLV. Each step of the original VSM method was evaluated to address the specific challenges of HMLV. Their work highlighted the importance of customer order decoupling point (CODP) as the crucial factor for the performance of ETO.

4.5. Material Allocation

The HMLV’s environment of uncertainties coupled with the increasing trend of outsourcing often has a wide supply base and highly random customer connection. This results in a large bill of materials with a variety of routings. This requires the consideration of customers’ demands and logistic factors in the material allocation system. Ali, et al. [49] addressed the concern with a material allocation architecture and genetic algorithm (GA) technique. By balancing customer-demand and committed demand for optimal resource allocation, their database system could be easily integrated with other platforms, such as MRP and ERP.

Line-level approaches to reducing energy consumption were usually focused on conditions where lot size is constant. The diverse customers’ needs and mass customisation in HMLV resulted in variations in lot sizes. Hibino, et al. [50] proposed a formulation to account for such variation with lot size management for optimal energy consumption per unit of production throughput for variable lot sizes.

4.6. Lean Transformation

As covered in the previous systematic literature review [4], lean manufacturing is another popular research topic in HMLV. Therefore, this paper will only cover the application of lean in the HMLV industry.

Conceptual models have been proposed for lean product development with practical applications in the HMLV business setting [51]. The adoption of kanban [52,53] and kaizen [54] in HMLV was studied in an actual industrial case study and showed positive results.

Industry-specific research on lean was mainly in electronic manufacturing and engineering support industries. Raghavan, Yoon and Srihari [53] designed and tested a lean framework for PCBA with the implementation of kanban for different lines and the implementation of a pull system. At the time, a single-minute exchange of die (SMED) philosophy was also being researched for electronics assembly [55], as well as in the engineering support industry [56,57]. The topic of research in lean concerns the implementation of the IoT for SMED procedures [55], a framework for analysing the introduction of automation [56], a framework for lean transformation [53], and before/after studies [54,57].

5. Models and Algorithms for HMLV Scheduling

5.1. Demand Management

Demand management in HMLV is challenging, with demand and product mix variability. Management of these demands is important to satisfy the order fulfilment. Looking into managing customer demand, Zhang and Tseng [58] observed that besides manufacturing flexibility, customers also have flexibility in their demands. They have a range of tolerance and sensitivity to certain product attributes and delivery schedules. Their developed model characterised this behaviour to integrate them into the HMLV order commitment model for optimising both interests. One consideration of this model is that the manufacturer must have full knowledge of their customers’ flexibility.

Meanwhile, Gadalla and Asme [59] developed an agility model for SMEs to improve their responsiveness to changes in customer requirements. It consists of four main enablers: quick manufacturing response strategies, multi-channel approach, HMLV techniques, and collaborative networks. With a use case tested on an SME business in the sheet metal industry, it highlighted the few key factors for a high level of agility: leadership, customisability of automated cells, quick response manufacturing (QRM) understanding, and supplier–customer ties.

For larger HMLV businesses with multiple manufacturing sites, Tan, et al. [60] proposed a genetic algorithm (GA) method to manage highly customised online orders by allocating them to different MTO production sites. The proposed formulation is based on the classical multi-knapsack problem to efficiently allocate customized orders across different MTO production sites while adhering to the constraint characteristics of typical job shop environments.

5.2. Production Scheduling

Production scheduling determines the resource allocations and job sequences to produce goods and services, usually by determining the start and end of each job on each machine. It must define its priorities and organise activities to meet the requirements, constraints or objectives [61].

Chou, et al. [62] investigated the effect of product variety under a stochastic job arrival and resource available for a job-machine assignment problem. The effect of job mixes under batching was analysed by constructing knowledge rules for reactive assignment for Poisson job arrivals and homogenous serial batch machines. They derived a few knowledge rules of dominance relationships between job mixings that could be used for reactive assignments.

Production levelling that is traditionally implemented in repetitive production environments with limited production diversity, was also adapted for HMLV production. Bohnen, et al. [63] used the principles and methods of group technology to achieve production levelling, according to a family-oriented pattern. Using unsupervised learning for different clustering algorithms, they quantified the product type similarity according to attributes specified by the grouping criteria. This resulted in different grouping results that were validated by the desirability index. The formed product families were then subjected to levelling patterns created by the specified EFEI-value (every part every interval) to achieve a minimal EFEI-value for cost-effectiveness.

Industry-Specific Scheduling Problem

In the electronic manufacturing industry’s PCBA, Chen and Chyu [64] formulated a binary linear programming model to optimise the best board assembly sequence on the PCB surface mount device machine (SMD) to arrive at the best setup strategy. Then a heuristic approach was proposed to improve the efficiency of finding the solution quickly.

While Salonen, et al. [65] work focused on group setup and minimum setup strategies. It evaluated the time of single component feeder changes that typically take 1–5 min, as well as a setup occasion where the component setup operations take 15–25 min. They formulated the hybrid machine setup problem, called job grouping with a minimal feeder setup (JGMFS) problem, as an integer programming model. Two hybrid algorithms were proposed, based on the efficient grouping and minimum setup heuristic.

Haiming, et al. [66] modelled the optimisation of the scheduling problem as an approximation of the traveling salesman problem. The scatter search model was used to develop an algorithm to determine the assembly sequence of different types of PCBs.

The recent development considered the practical constraints of the surface mount device and the technology [67]. PCB characteristics and configuration in the surface mount technology production line were considered, such as product type, PCB board type, the number of points, dimensions, and SMD configurations. There, a hybrid symbiotic organism search (SOS) algorithm-based support vector regression (SVR) model was proposed to estimate the throughput in flexible, complex, and uncertain PCBA processes for production scheduling.

In the semiconductor industry, wafer fabrication is carried out on multiple processing modules. The wafer has strict residency constraints where the wafer’s surface would be detrimental if it were to stay at a processing module for too long after being processed. Taking this wafer residency time as a constraint, Zhu, et al. [68] presented a Petri net model for the close-down process of a single-arm cluster tool.

Since production in the semiconductor industry is automated, its research involves optimally scheduling the transient processing of single-arm multi-cluster tool robot [68,69]. Mhiri, et al. [70] also specifically investigated finite capacity planning by proposing an algorithm to project production loss trajectories; this determined the expected load for all machines and balanced the workload against bottlenecks. The team also came up with two planning approaches, mixed integer program and WIP projection, with the former being the better solution [71].

Similarly, perishability was considered in the scheduling problem in the food processing industry. Matsumoto, et al. [72] aimed to find the optimal sequence of ingredients boiling in a jam plant. The setup time in changing products in the boiling process was a production bottleneck. Then, the sum of earliness and tardiness was reduced by the proposed modelling, based on the Lagrangian relaxation.

Meanwhile, in the precision engineering industry with manual operation, workforce scheduling includes the factor of the number of employees, legal work limits, and company requirements. Given that operators have different skill sets and job grades, Pan et al. (2010) consider operator costs and staff satisfaction as one of their scheduling methods. Their work uses mixed-integer programming with a two-stage heuristic algorithm to solve the workforce scheduling problem in the HMLV precision engineering job shop manufacturing environment.

In die-casting manufacturing, the casting parts can be moulded conveniently with few limitations in the product’s shape and size. The die-cast itself has many process variables that can affect its product quality. Therefore, careful calculation of scheduling is important in a multi-shift casting environment. Based on the real constraints, Park and Yang [73] have formulated a linear programming model to solve the scheduling problem and optimise the furnace’s efficiency.

However, it is noted that only one study addresses the consideration for the workers, such as workers’ skill requirements for each shift and operators’ work preferences [74]. The author proposed a programming model which schedules the workers’ shifts according to their skill sets and work preferences.

5.3. Decision Support System

For good production scheduling flexibility, the predictive-reactive scheduling approaches were adopted by the HMLV industry. Whereby the offline production schedule generated before the production was corrected by the online production schedule during the production [61].

Evaluation of the HMLV decision is an important aspect of production planning. Earlier research was carried out on algorithm automating operation planning by extracting machining methods for a numerical control (NC) program. The devised system automatically sequences the optimal work order with the correct tool [75]. Later work focused on mathematical models and simulation to forecast manufacturing uncertainties (machine, labour, and logistics) [76], defects in production [77], and demand forecasting with consideration of internal and external factors [78].

The advanced machine learning methods in model identification and behaviour prediction of the complex production system enable DES as an efficient tool for analysing the complex impact of decisions made concerning both static structure and dynamic behaviour of a production system. However, they hardly exploit the engineering background knowledge available in the production system [79].

Real-time data recorded by sensors can be used for smart scheduling to achieve real-time, reliable scheduling that can be integrated with workflow in real-time. Industry 4.0 concept of cyber-physical systems (CPS), which collects the real factories with big data, offers new opportunities for process planning.

Recent research trends are in the development of a comprehensive decision support system. Gödri, Kardos, Pfeiffer and Váncza [79] used model- and data-driven analyses and unsupervised machine learning to reduce the input domain size required for comprehensive analysis. Perraudat, et al. [80] developed a system to support critical time-constrained operations in wafer fabrication. While [61] used the multi-broker genetic algorithm (MB-GA) in scheduling, with an advanced and effective real-time production scheduling decision support system model.

5.4. Type of Common Modelling and Simulation Methods

5.4.1. Common Algorithms and Programming

Scheduling for HMLV focuses on optimising the challenges that arise from high product variation. Mathematical modelling, simulations, and algorithms were developed considering the different parameters and policies.

Heuristic algorithms were developed to reduce the computational time and provide efficient evaluation methods for their problems [81]. The genetic algorithm (GA) was the most popular metaheuristic algorithm.

Table 4. List of algorithms used and their objectives.

Heuristic Algorithm
W. S. Chen and Chyu (2003) [64]	Two-phase searching algorithm to solve binary integer problem—PCBA job sequencing
Haiming et al. (2009) [66]	Traveling salesman problem algorithm—production sequence
Pan et al. (2010) [74]	Mixed integer programming model with two stage heuristic algorithm—manpower scheduling
Srinivasan and Viswanathan (2010) [81]	Heuristics model for closed queuing network (CQN)—WIP optimisation
Matsumoto et al. (2011) [72]	Heuristic method based on the Lagrangian relaxation—scheduling perishables
Svancara and Kralova (2012) [82]	Job shop scheduling
Arasanipalai Raghavan et al. (2015) [83]	Shortest total estimated processing time (STEPT)—scheduling stochastic rework
Ojstersek and Buchmeister (2020) [84]	Kalman algorithm—flexible manufacturing simulation
Genetic Algorithm
A. Ali et al. (2010) [49]	Material allocation
Neoh et al. (2010) [85]	Product mix optimisation—GA based TOC; particle swarm optimisation; layered-encoding cascade optimisation
Gissrau and Rose (2012) [86]	Dispatching policy
Fabian Bohnen et al. (2013) [87]	Levelling with group technology—non-dominated sorting genetic algorithm II (NSGA-II)
F. Chen et al. (2014) [88]	Job allocation—resource-constrained project scheduling problems (RCPSPs)
Ohta et al. (2014) [78]	Demand forecasting
C. S. Tan et al. (2018) [60]	Order allocation to MTO production sites
Linear Programming
Park and Yang (2013) [73]	LP for maximum molten alloy usage—die cast scheduling
Q. Zhu et al. (2017) [69]	Optimal schedule for transient processes—RPA scheduling
Alduaij and Hassan (2020) [40]	Machine distance algorithm—robotic layout
Reinforced Learning
Zheng et al. (2020) [89]	Deep learning with transfer approach—dispatching policy
M. Feng and Y. Li (2022) [90]	Predictive maintenance decision model—Markov chain model with dynamic programming
Zhou et al. (2022) [91]	Online scheduling for decision making—Markov decision process (MDP)

shows the algorithms found and their main objectives.

5.4.2. Discrete-Event Simulation

DESs were the common methodology to simulate the HMLV production system.

Table 5. List of DES-related work and their objectives.

Discrete-Event Simulation
J. Li et al. (2011) [92]	Factory modelling for PCBA
Lange et al. (2012) [42]	Order fulfilment based on virtual time-based flow principle
Gödri et al. (2019) [79]	Simulation-based decision support
Khan and Standridge (2019) [43]	Setting CONWIP limit
Robert et al. (2019) [93]	Visualisation of simulation on VR
Okubo and Mitsuyuki (2021) [94]	Backward scheduling
WITNESS
Švančara (2011) [95]	Linking database with witness for scheduling
Eng et al. (2015) [18]	POLCA authorisation simulation
Hibino et al. (2019) [50]	Energy consumption based on lot-size optimisation
ARENA
Zhang and Tseng (2009) [58]	Order fulfilment model based on customer’s flexibility
Fernandes et al. (2019) [46]	POLCA—centralised vs decentralised job release
Alnahhal et al. (2021) [96]	Modelling freight—temporal consolidation for MTO supply chain
AnyLogic
Gissrau and Rose (2011) [97]	Fab simulation model
SimPy
Thürer et al. (2020) [12]	WLC assessment in additive manufacturing ORR
FlexSim
A. Miqueo et al. (2022) [98]	Production modelling of manual or semi-automated assembly lines

shows some of the specific software being noted.

5.4.3. Validation of Modelling and Simulations in the Industry

The extent of the validation of mathematical modelling and algorithms that were tested or validated, shown in Figure 11, highlights the lack of actual industrial validation in areas other than production and operation management, mainly production planning and scheduling. The electronic and semiconductor industries were the main ones where validation work was carried out.

6. Automation and Robotic Production in HMLV Production

The need to cope with the demand for mass customisation and market challenges is pivoting towards extending the levels of industrial automation while being cost-effective and flexible [13,23]. This chapter explores the research on automation, robotics, and human–robot collaboration for the HMLV.

6.1. Robotics in HMLV Manufacturing Systems

Robotics and automation in production systems were well covered in other literature reviews. Therefore this review covers the interest of robotics in production where HMLV were addressed.

6.1.1. Flexible Batch Manufacturing

Batch manufacturing systems are one of the manufacturing strategies for HMLV. The research addresses multi-robot coordination in flexible batch manufacturing systems to solve its bottleneck issue with developed algorithms and simulations [99]. Meanwhile, working on the best robotic team configuration, Liemhetcharat and Veloso [100] devised a synergy graph method to compose teams by selecting modules to fit the required tasks best.

6.1.2. Programming and Algorithm Training

A high variation of products meant that robots had to be programmed frequently. New methods and approaches are used to minimise the effort of setting up the robotic system. One of the most prominent ways of achieving this goal is to minimise the need for reprogramming to reduce the costs of employing robots, especially for SMEs [101].

Traditionally, robot teaching involves recording the waypoints of the robot’s end-effector, and then the path is interpolated to generate trajectories. Recording the waypoints is time-consuming and does not have the adaptability for complex trajectories or geometries, since it is difficult to collect multiple waypoints via pure manual teaching [102].

One of the lead-through ways is where the user manipulates the robot along the desired path to the desired target points, where the robot records the entire process. This form of learning by demonstration (LfD) is an easy way to program the robots intuitively. Ko, et al. [103] expanded on this method by developing an algorithm to analyse segments of the operations. Then, human operators can perform corrections and evaluations to optimise the operations. Kito, et al. [104] also optimised the training method by including the robot’s visual feedback and force sensors.

On the contrary, programming can be carried out without the actual robot, this is called offline programming. This method is performed virtually with the robot and its workplace in a virtual environment. The teaching of the robotic tasks is achieved on the virtual robot. Therefore, programmers can teach the robots without interrupting production processes, thus reducing the unnecessary downtime of the robots [105]. Both teams, Ong et al. (2020) and Solyman et al. (2020), developed augmented reality systems where users can move around the work cell to define the target point and path the robot should follow.

Telemanipulation is an offline programming system where the operators guide robotic systems online, with interactive devices, due to the large flexibility from the high cognitive and perceptual capabilities of the human-in-the-loop robot teaching process. Telemanipulation’s limitations include portability, low accuracy, and cumbersome motion mapping methods. These limit the effective robot teaching in multi-degree-of-freedom (DoF) manipulation under force and path constraints.

Therefore, more straightforward and efficient solutions have been developed for telemanipulation. One such use case in the HMLV industry is in the aero-manufacturing industry. Tape masking is used to cover objects’ surfaces during industrial processes, such as plasma spraying, spray painting, and shot spending. Work is being carried out to improve the learning capability of the path that is taught by human telemanipulation to provide automation solutions, to free operators from these tedious and unhealthy workloads [102,105].

To improve learning in actions that cannot be optimised due to unknown parameters, Sorensen, et al. [106] developed a method with Bayesian optimisation for a generic iterative approach in robotic learning. The users set the correct outcome of the action (e.g., dropping a ring into a fixture), and a camera captures the desired outcome. Then, the robot iterates different parameters (distance, height, and angle) of the operation until the desired result (from the camera) is achieved. Therefore, the robot can automatically find the most reliable solution for the parameters.

6.2. Human–Robot Collaboration (HRC)

Human–robot collaboration (HRC) bridges the advantages of human operators and robotics by combining the robot’s precision, repeatability, and strength with human intelligence and flexibility under variable conditions [107]. However, the safety of humans working in proximity to the robots and task allocation between the human–robot are the two main concerns of HRC for HMLV manufacturing.

6.2.1. Task Allocation

The efficient usage of collaborative robots has become a key role in the dynamic environments of the HMLV industry. To harness both the advantage of the operators and robots, Malik and Bilberg [13] devised a method to evaluate the complexity of the assembly operations and distributed the task accordingly between humans and robots with consideration of the physical properties, methodology and safety. Similarly, Chen, Sekiyama, Cannella and Fukuda [88] devised an algorithm to determine the suitable sub-task allocation strategy between humans and robots to maximise the effectiveness between assembly time and costs.

6.2.2. Safety and Trust

To improve the safety of humans working together with collaborative robots, Chen, et al. [108] developed an algorithm that estimated human intention and determined the time when the task was transferred from human to robot. Another method of estimating human and robot activities and intentions has been proposed by Cramer, Cramer, Kellens and Demeester [107]. It looked at object affordances—the interdependencies among objects, motions, and human activities, to develop an algorithm that predicts the actions of humans.

To improve the trust of humans towards collaborative robots, Palmarini, et al. [109] designed an AR interface that provided situational awareness and spatial dialogue while working with a collaborative robot. As the operator performs the operation, they can see the robot’s operation in advance.

7. Manual Production Optimised for HMLV

The manual assembly line is still an economical and reliable manufacturing system to respond to the HMLV market trend. Although an automated and robotic production, its costs are a prohibiting drawback for SMEs for assembly and inspection [101,110]. Large manufacturers still depend on manual assembly in their production areas as setup and maintenance costs for automated systems are ineffective. This section highlights the manual assembly research, explicitly addressing the HMLV industries.

Not many research studies addressing manual assembly work in HMLV production have been carried out, with the objectives of improving the working ergonomics and increasing productivity. In 2003, Quintana and Hernandez-Masser [110] investigated manual electronics assembly to draw up a design criteria framework for the assembly line. They identified limiting design criteria of human ergonomics that affect productivity through biomechanical and throughput data analyses. From there, a framework was established as a guideline for general workstation design.

7.1. Consideration of Assembly Complexity

As the variation of assembled products increases, the assembly complexity increases with the number and diversity of the parts and fasteners in the product assembly. This creates extra effort for operators to recognise, grasp, orient, insert, and assemble the product [111]. Therefore, the objective evaluation of assembly complexity becomes the first of many steps for manufacturers to stay competitive.

Quantifiable metrics were created to measure the assembly complexity. These methods can be broadly based on objective and subjective measures [112]. For example, Zeltzer, et al. [113] complexity calculator (CXC) collects objective data, including the number of different packaging, assembly directions, unique parts in the workstation, and the number of tools.

Subjective measures of complexity rely on feedback from the operators and evaluators. From the empirical data obtained from workshops and interviews [114,115], the complete complexity of the production system can be assessed in the aspects of (i) product variants, (ii) work content, (iii) layout of the workstation, (iv) tools and equipment, (v) work instructions, and (vi) general opinion of the work station [112].

From the systematic literature search, Falck, et al. [116] developed an assembly complexity assessment model with huge model variations for car assembly. It examined the significance of assembly complexity with the assembly ergonomics and quality failures in car manufacturing. Their feedback on assembly ergonomics and complexity was collected by interviewing engineering and operators. With this, the developed basic complexity criteria (CXB) use 16 criteria to identify high-complexity assemblies. The authors noted that each complexity criterion’s judgement is harder to pinpoint on a scale. Therefore, the CXB method uses a direct yes or no answer scheme, which makes the assessment easier. By applying the study in car assembly of 47 different assembly tasks, covering 47,000 cars, the research found that high-complexity tasks significantly affect the failure output, increasing its action costs.

7.2. The Assembly Assistance System

Industry 4.0 brings new interactions between operators and machines, which will empower the intelligent workforce to significantly affect the nature of their work. Operator 4.0 is a concept of human-cyber-physical systems (H-CPSs) that facilitate cooperation between humans and machines powered by technologies, such as smart sensors, IoT architectures, big data analytics, and augmented reality [117].

The Operator 4.0 concept aims to improve the abilities of the operators. With adaptive systems with real-time support and performance monitoring, all of the data related to operator activities are measured, converted, analysed, and transformed into actionable knowledge and presented to the operators [118]. These forms of adaptive systems under the Operator 4.0 concept can be assembly assistance systems for the operators working in manual assembly in HMLV manufacturing.

As HMLV has increased product complexity and assembly complexity, research is being carried out to create assistants to help with the increasing cognitive mental load [25]. These assembly assistance systems support the operators to cope with the high variety of information exchange by providing information timely and accurately to minimise the cognitive workload in understanding the information where operators can make analyses and informed decisions in the flexible manufacturing system.

Some of these assembly assistance systems utilise AR technologies to provide visual assistance to the operators. One such system, from the literature review, was developed by Verhoosel and van Bekkum [119]. A spatial AR system was used to project information onto the workstation surface, where the assembly process was presented to the operators. The system was also connected to the MES and ERP to relay product design and configuration to the operator.

Assembly assistance systems are also useful for training purposes and guiding novice operators. AR user testing has shown its effectiveness in supporting training for novice operators in industrial settings. They provide high time efficiency, clear instructions, and the ability to achieve the immediate result of a finished product [24].

8. Future Research Direction

8.1. Production and Operation Management

8.1.1. Industrial Testing and Application

Industrial testing of the research work remains the most important future work that needs to be carried out to validate the proposed solutions. This is to test the algorithm under real-life problems and investigate them in uncertain and dynamic conditions, based on actual industrial demands [40,91,120,121].

The importance of the evaluation, based on complex real-world HMLV conditions was highlighted by several authors [18,46,47,91,122]. This is to account for the broader range of manufacturing environments and production conditions.

8.1.2. Comprehensive Decision Support System

The topic concerns better mathematical modelling, simulation, and algorithm for decision-making. Some work was found to be specific in solving a particular problem (e.g., scheduling and sequencing or job release). However, with the complications of the HMLV industry, multi-aspect decision support systems that cover as much of the value-adding process are crucial to aid the complex multi-factor decision-making process in HMLV production.

Of the decision support systems found from the review, all showed positive results from simulated real-world data [61,78,80]. Of the research carried out, very little has been integrated and implemented in real-world applications [13,79].

There is very little literature on integrating the simulation models with a digital twin. Communication between simulation software and the real factory would open new opportunities for a continuous correction capability to enable self-optimisation and self-learning [79].

Storing and sharing experts’ knowledge has been achieved with systems, such as the advanced documentation and control system (ACDS) [123]. Such a system stores experts’ input on maintenance procedures and advanced process controls (APCs). However, further integration of expert knowledge into the decision support system helps to optimise the algorithm, while the machine learning algorithm can complement experts’ opinions [124]. Experts can make adequate decisions with their specialised techniques and deal with expected events by finding general rules from accidental events [72].

Collecting, evaluating, and integrating experts’ knowledge into the decision support system has been achieved by past researchers. Future work could investigate filtering and balancing the objective technical input of experts versus experts’ personal judgement [96].

8.1.3. Industry 4.0 Integration

While Industry 4.0 has widely been researched in the manufacturing domain, the discussion of Industry 4.0 with HMLV-specific problems has been rare. In ICT and IoT, most of the work found was published after 2018, except for one [55]. Their work concerns the implementation of IoT devices in production processes.

Another prominent Industry 4.0 technology research was AR and VR [93,101,119]. These works improved human–robot collaborations with better training and safer programming and increased the operator’s productivity through more efficient information management.

As costs are essential to staying competitive in this industry, implementing Industry 4.0 technologies for SMEs is another paradigm to be explored as there is a demand for cost-effective solutions in this pressing time of digitalisation.

Studies on the effects of digitalisation and their evaluation tools can help practitioners to understand how the effects of the investments can be maximised to achieve their desired operational or business goals. These could involve comparing the estimated improvements against the costs of implementing Industry 4.0 technologies with actual results in the industrial environment [98]. Further research should be carried out on investigating the effect of Industry 4.0 adoption through a comprehensive model to improve the production and sustainability measures [125].

Similarly, the Industry 4.0 adoption model in the HMLV manufacturing industry could explore the adoption factors in the market for decision-makers [126]. To arrive at a framework for Industry 4.0 digitalisation and transformation that suits different sizes and categories of HMLV businesses can also be explored. This could harness the advantages of the different types of manufacturing companies with various levels of human–machine combinations.

8.2. Other Areas of Production

Research in HMLV manufacturing could also expand beyond production planning and management. Other areas of production remain underexplored, such as downstream processes and operations. This is to examine the constraints of supply processes, supply shortages, and breakdown [37]. For instance, Lee, et al. [127] evaluated the design and effectiveness of the IoT-based warehouse inventory management system for HMLV manufacturers to maximise the performance of the receiving, storage, and picking activities in the warehouse. Incorporating AI would be the future direction to enable smart logistics, information automation, and warehouse streamlining for better efficiency, performance, and costs.

8.3. Energy Efficiency and Sustainability

The climate change crisis presses for net zero carbon strategies to reduce carbon emissions [128]. Industry 4.0 technologies could potentially address traditional practices and technologies’ ecological and social limitations. This would improve long-term organisational competitiveness.

Meanwhile, the rising concern about energy costs prioritises the need for renewable energy, energy conservation and efficiency as effective options for sustainable energy solutions. Tan, et al. [129] energy efficiency benchmarking methodology has shown the need for effective energy efficiency benchmarking to identify the best action plans and reachable energy efficiency goals within a company or across the industrial sector.

Sustainability benchmarking can be expanded into the simulation study of the HMLV, such as the flexible manufacturing system [130]. For instance, the study on the integration of robotics for HRC in the FMS was from the cost-time investment and production sustainability perspective. This could be extended to evaluating the environmental sustainability of a highly flexible human–robotic collaboration in HMLV.

Sustainability and energy efficiency has been integrated into the decision model for joint production and maintenance decision-making [90]. Future work could optimise production and maintenance to factor machines with continuous health states where machines could be configured to be temporarily turned off for energy saving by optimisation of the energy saving control without inflicting production loss.

8.4. Automation and Robotic Production

8.4.1. Mix and Hybrid Assembly System

The HMLV industry still utilises a mixed assembly system with automation, robotics and human operators for a mix of automated, semi-automated, and manual production systems [13,88]. Most of the research has been focused on planning and implementing such a system. In this aspect, future research could focus on the real-world case studies on the effect of mixed or hybrid assembly systems on safety, ergonomics, performance, and sustainability.

8.4.2. Programming

Programming the robot for production automation has been the key challenge of implementing robotics in HMLV production. From the research carried out on the improvement to ease the programming of robots [101,104,106,131], future research needs to make programming and the configuration of robots easier and cost-effective, such as kinaesthetic teaching [106]. In contrast with typical production automation at a low-volume, high-mix environment, product variations require frequent robot setups and re-configurations, which require skilled programmers and technicians. Therefore, making the programming of robots easier with minimum skills, minimising setup procedures, and re-configuration costs would decrease the cost of robotics and automation for HMLV.

8.4.3. Human–Robot Collaboration (HRC)

Beyond the evaluation of mixed assembly systems, HRC research has focused on the working of humans with robots. The research has focused on two critical aspects of HRC: ergonomics and safety.

Ergonomics could include the physical and cognitive ergonomics of working with cobots in an HMLV environment. Physical ergonomics concerns better task scheduling, motion planning, and control to minimise physical stress through workload sharing and scheduling [22]. At the same time, cognitive ergonomics could look at the psychosocial risks of humans working with robots, especially with the variability of the workflow.

In terms of safety, research in robotics mainly concerns two elements: contact avoidance and contact detection and mitigation. As complexities increases with product mix, one such example of research on HRC safety in HMLV is the mathematical model to predict each other’s intentions [107]. Therefore, there is an opportunity in the safety domain in HRC to increase humans’ trust in working with robots.

8.5. Manual Assembly

As with HRC, improving manual assembly in HMLV centres around performance optimisation in physical and cognitive ergonomics. Although the topic is well-researched, the factor of variability in workers’ workflow could be explored in the domain of physical ergonomics in manual assembly.

Assembly assistance systems, or operator support systems, are one of the proposed solutions to improve the operator’s flexibility for HMLV. These systems were mainly prototypes and saw limited testing and application in the real industry. Future research could evaluate their effectiveness in the industry in terms of performance, practicality, user acceptance, and costs. Integration of such systems into the industry-wide IT systems is another dimension of research to maximise their functionality under the concept of Industry 4.0—Operator 4.0.

9. Discussion

This section highlights the research questions answered from the previous sections, organizing them as answers to the seven research questions posed earlier.

9.1. Past Trends in Two Decades and Current Area of Interest

Question 1: “What has been the trend of HMLV research for the past two decades?”.

By charting the literature found according to the industrial sectors, 71% of the research was not industry specific. The industries with prominent HMLV research problems were the semiconductor, electronics, and engineering support industries. Regarding the research area, 63% were about production and operation management. This showed the heavy emphasis on production scheduling and production planning. Therefore, there are opportunities for research in the other aspects of the value chain.

Question 2: “What are the current areas of interest in HMLV research?”.

From 2017 to 2022, Industry 4.0 technologies have seen growing interest in topics, such as robotics, artificial intelligence, the internet of things, and additive manufacturing. Beyond algorithms and mathematical models for production planning, the concept of big data advocates production data collection to allow comprehensive data analysis for the optimisation of decision-making systems.

9.2. The Application of Digital Tools and The Extent of Simulation Testing in the Industry

Question 3: “How are digital tools being used in HMLV for production management?”

The production planning work has progressed over two decades from mathematical modelling and algorithms. The new Industry 4.0 concept advances towards using decision support systems for the production planning of HMLV. Digital twin and big data enabled data collection and reactive planning, based on real-time factory data. At the same time, algorithms can self-learn and optimise reactively, increasing the flexibility of the decision-making process against variations and changes.

More real-world industrial user testing is expected in coming years to include comprehensive evaluations of their developed work. This is a continuation of the simulation and mathematical research for industrial validation.

Question 4: “What is the extent of simulation being tested on the industry?”

Categorisation of the industrial validations, with industrial testing and dataset, carried out by the literature, according to the industrial sector. Most of the validations were on production and operation management solutions for production planning. As the electronic and semiconductor industry were the two dominating industries for validations, this highlighted the gap in research in other HMLV manufacturing industries.

9.3. Automation and Robotic Production in HMLV

Question 5: “How has research in automation and robotic production contributed to HMLV?”

The demands for mass customisation and market challenges have driven manufacturers to extend the levels of automation in HMLV production while being cost-effective and flexible. Flexible batch manufacturing is one of the strategies for HMLV, which aims to use multi-robot setups to suit the variation of tasks. New programming methods have also minimised the setup costs of the robotic system. Techniques, such as learning by demonstration, offline programming, and telemanipulation have shortened the robots’ programming time and downtime. The robotics capability is also actively being researched to explore more possibilities to automate monotonous, dangerous, and un-ergonomic assembly steps.

Human and robotic collaboration combines the advantages of human operators with robotics. Robots have good precision, repeatability, and strength, while human intelligence and flexibility are valuable under variable conditions. Work has been carried out to optimise the task allocation between the two with consideration of the task properties. Safety and trust of the worker working alongside the collaborative robots are also key concerns addressed by the research work.

9.4. Manual Production in HMLV

Question 6: “How is manual production optimised for HMLV?”

Few studies address ergonomics and productivity in the HMLV environment. The assembly complexity assessment model has been developed to measure the complexity of the assembly tasks to provide the foundation of work allocation to match workers’ suitability.

Assembly assistance systems, in the form of digital work instruction, have also been developed to improve the operators’ abilities of the operators through the concept of Operator 4.0. Operator 4.0 includes assistance systems with adaptive real-time support, performance monitoring and data collection to aid the operator in manual assembly. Such assistance systems hope to reduce the cognitive load of the operators to cope with the high variety of information exchange while working in an HMLV environment.

9.5. Future Direction of HMLV Research

Question 7: “What is the future direction for HMLV research?”

While validations were carried out on generated data, there are opportunities for industrial-specific validations. Research in production areas downstream of the value chain could also be explored, such as logistics and product lifecycle management. Long-term evaluation of the applications of production technologies, such as cobots, AR, and VR, is another area to be explored.

Climate change and the energy crisis made energy efficiency and sustainability the most important research agenda across all industries. There is room for optimisation of all aspects of the value chain, which could be enabled with Industry 4.0.

The future of HMLV would need to work on integrating data and supply chain processes, ultimately arriving at an advanced decision support system, as touted in the concept of Industry 4.0. This would deliver greater flexibility beyond the manufacturing system, including supply chain partners. Eventually, the research for HMLV would extend beyond production design and management theory. It would increasingly cover the popular topic of Industry 4.0—its concept and new production technology—to offer greater manufacturing flexibility.

10. Conclusions

This paper reviews the HMLV manufacturing industry from the industry perspective to reveal the research that addresses the HMLV production issues, categorising them according to their field of improvement and the industrial sectors. As such, 152 documents were analysed. The resulting analysis was structured into seven main sections to address the seven questions posed in this paper.

It was found that most of the research centres around the production planning for HMLV manufacturing without being industry-specific. The semiconductor and electronics industry are two prominent industries with the most interest vested in the research and validation of the work. This highlighted the research opportunities for practical research and applications in other HMLV industries. Industry 4.0 technologies brought new possibilities to manufacturing with big data, the Internet of things, robotics and automation, and decision-support systems. However, explicit discussion of these technologies for industry-specific HMLV manufacturing is still in need. This includes the adoption model of new technologies to evaluate the costs and potential benefits. Meanwhile, the pressing need for decarbonisation makes sustainability an ever-important factor in production planning and technological adoption in the HMLV industry.

The limitation of this review is that this paper only includes literature with keywords specifically mentioning HMLV manufacturing in writing. Therefore research work and technologies that have or may be applied to HMLV could be missed from this review process. Furthermore, the data also covered the existing work without surveying the factors of the adoption of Industry 4.0 technologies in HMLV.

Therefore, there is room for future work investigating the factors of technological adoption in HMLV. This could consider the company-wide digitalization of the HMLV enterprises or delve into the adoption of individual technologies for the HMLV industries. This includes investigating the key process parameters of HMLV manufacturing for adopting Industry 4.0 technologies.