Making electric vehicle batteries safer through better inspection using artificial intelligence and cobots

High quality, safe, and reliable batteries are essential for widespread adoption of electric vehicles. Current Li-ion battery pack manufacturing processes rely on manual inspections to ensure electric vehicle battery quality. Such manual quality control is prone to errors, increasing the chances of defective batteries. This is likely to increase safety and reliability concerns in the public imagination, slowing down the adoption of electric vehicles. Furthermore, manual inspection is time-consuming and likely to become a bottleneck in scaling up electric vehicle battery production. A potential solution to address this need for fast and accurate inspection of batteries is the use of machine vision and robotics. In this study, we use digital twin design and simulation to develop a battery module inspection system that uses cobots and machine vision to inspect electric vehicle batteries for defects. Our proposed system can automate visual quality checks that are currently being done by human operators. The proposed cobotic system has been simulated and validated for a variety of battery defects to achieve fast and reliable detection. Since a digital twin of the cobotic inspection workcell has been used, the battery inspection system, as designed and validated, is ready for immediate implementation.


EV battery safety -critical for EV adoption
As the adoption of electric vehicles (EVs) has increased, so has the frequency of safety incidents associated with them.A large share of these incidents are related to the EV battery.Faulty battery systems can cause on-road power failures, fires, or even explosions (Lawrence 2022).In a watershed moment for the automotive industry, GM recalled the Chevy Bolt resulting in losses of over USD 1 billion (Isidore 2021).When announcing the recall, GM stated: "After further investigation into the manufacturing processes at LG and disassembling battery packs, GM discovered manufacturing defects in certain battery cells produced at LG manufacturing facilities beyond the Ochang, Korea, plant.GM and LG are working to rectify the cause of these defects.. . .In rare circumstances, the batteries supplied to GM for these vehicles may have two manufacturing defectsa torn anode tab and folded separator present in the same battery cell, which increases the risk of fire.. . ." - (GM 2021) CONTACT Ajit Sharma ajit.sharma@wayne.eduSupplemental data for this article can be accessed online at https://doi.org/10.1080/00207543.2023.2180308.
Although the incidence of battery-related defects is infrequent, such incidents may pose significant cost and reputational risk for EV firms.Additionally, battery packs are a new component in the automobile, further amplifying the fear associated with unknown safety hazards in consumers.Such public anxiety about battery safety can mute the adoption of EV vehicles.Therefore, it is critical to ensure the safety of EV batteries for the widespread adoption of EVs as a way to transition to a future of zero emissions.

Reducing manufacturing defects -a key to reducing EV battery safety incidents
Battery defects can be induced in the EV battery due to three broad categories of factors: (1) external triggers, (2) degradation of battery quality with time and usage, and (3) defects induced during the manufacturing process (Figure 1).These three broad categories emerged from a study of academic literature (Kaliaperumal et al. 2021;Hendricks et al. 2015;Palacín and de Guibert 2016) as well as other sources such as white papers, websites of vendors of battery testing equipment and services, and advertisements in trade publications.The first category of factors, external triggers, are chance events such as accidents or collisions leading to a puncture and subsequent leakage of electrolyte which can prove to be hazardous (Yoo Jung Park et al. 2018).While EVs can be designed to mitigate the impact of such events, their occurrence is not within the control of the EV manufacturer.Similarly, the second category of factors are also out of the control of the EV manufacturer.The passage of time is a given that cannot be managed or controlled.The level at which the battery gets used is a variable factor.Some owners may drive the EV at a significantly higher level than others who use it occasionally.Similarly, there can be a wide spectrum in the level of care with which customers use the EV.For instance, users leaving the battery to charge constantly on a charger or using an incorrect charger, can cause the battery to age faster (Jialong Liu et al. 2020) as well as cause thermal runaways (Jialong Liu, Wang, and Bai 2022).Finally, EVs may be used in different climatic conditions which can lead to wide variability in the operating conditions.For instance, regions with high humidity can lead to moisture entering the battery pack which is one of the most common causes for battery swelling (Beck 2021).Swollen batteries can be extremely dangerous as they may cause the pack housing to get punctured, resulting in leakage of hazardous gases.The battery could also catch fire and explode, leading to serious injury.Similarly, charging a battery back in cold temperatures (below 32 • F) can cause lithium plating, which is the deposition of metallic lithium on anode (Lin et al. 2021).This plating is permanent, cannot be removed, and can lead to short circuits and easier damage from impacts.While all these factors are variable, they depend on the geographical location, driving frequency and driving behaviour of the customer, and are therefore out of control of the manufacturer.
In contrast, the third category of factors, i.e. manufacturing and material related factors, are within the control of the battery manufacturer.Manufacturers can take quality control measures to ensure that the material inputs are of a high grade.Similarly, manufacturers can take a wide variety of measures to minimise the defects that get induced into the battery during the battery manufacturing and assembly process.
According to research from IDTechEx (Edmondson 2021), a large share of battery fires occurred when the vehicle was parked, not driving or charging.Such instances highlight that the LIB manufacturing process can introduce defects that remain undetected and can cause a safety event during normal operation, without the occurrence of an additional external trigger.Therefore, it becomes imperative for the automotive industry to explore ways to reduce manufacturing defects in LIBs.This is the focus of this study as highlighted in Figure 1.

Can AI and robotics reduce manufacturing induced battery defects?
A search for publications on the use of AI and robotics for automated inspection in manufacturing reveals an accelerating uptrend in the use of robotics and machine vision for automated inspection (Sharma, Zhang, and Rai 2021;Nti et al. n.d;Smith, Smith, and Hansen 2021;Bhatt et al. 2021).This trend spans industries including manufacturing (Mok, Kwong, and Lau 2000;Buer, Strandhagen, and Chan 2018;Choudhary, Harding, and Tiwari 2009), aircraft manufacturing (Tsuzuki 2022), garment manufacturing (Abd Jelil 2018), food (Dora et al. 2021), and beverage manufacturing (Rahman et al. 2019).
The automotive industry also uses robotics extensively (Boudella, Sahin, and Dallery 2018;Chen et al. 2022;IFR 2021), which in part explains the notable improvements in product quality, as reflected in a declining 100PP metric (Figure E1).Given the increasing intensity of robotic and AI-based inspection in manufacturing (Chuqiao Xu et al. 2022;Perng, Lee, and Chou 2010;Jiang, Wang, and Hsu 2007;Häcker, Engelhardt, and Freyw 2002), and the urgent need to improve the quality and safety of the electric vehicle battery, it becomes imperative to ask the following question:

Can advances in AI and robotics help improve the quality of EV batteries?
To address this question, we must, in turn, answer the following questions: To answer these questions, it is important to start with a clear view of the state of current research and practice on the use of AI and robotics to reduce manufacturing defects in EV batteries.Our review reveals that research in this area is scant and scattered, focussing primarily on inspection in the upstream process of EV battery cell production.To the best of our knowledge, no study to date has studied automated inspection of EV battery module inspection (EVBM, hereafter).While the EV battery manufacturing process remains a heavily guarded secret, the popular press and trade publications point to manual inspection of EVBMs which is time consuming, tedious, and error-prone (Matousek 2018).Therefore, there is an urgent need to study the automated inspection of EVBMs.
In this study, we address this need by using the digital twin method to design, simulate, and validate an automated robotic inspection system to detect defects in EVBMs.Our proposed inspection system uses machine vision for automated detection of EVBM defects.The system was simulated to detect some of the most common EVBM defects, such as missing cells, missing or defective cell connections, missing terminals, correct cell polarity, and total cell count.AI error proofing techniques were used, which accurately detected defects in a fraction of the time required by human operators.In sum, our analysis found that the inspection of EVBMs is highly amenable to automation, with the exception of manual inspection steps for some types of defects.Specifically, checks such as snugness of fit of wires that require the dexterity, versatility, and sensitivity of the human hand, as well as coordination of multiple senses, remain difficult to automate.We have solved for this problem in our proposed system by using a cobot instead of a robot, allowing concurrent inspection of difficult to automate checks by a human operator.
Although it is possible to evaluate robotic automation solutions by setting up physical prototype work-cells, it can be an expensive and time consuming proposition.Our approach of using digital twin design and simulation allowed us to design, programme and simulate a robotic work-cell, taking into consideration real-world requirements and constraints, without the need and expense of setting up a physical prototype work-cell.This allowed us to simulate the exact workings of the inspection solution, providing insights into its performance, enabling fast iterations, and saving time.Because digital twin methods were used, the EVBM inspection system, as designed and validated, is ready for immediate implementation in the real world.Although EV battery safety remains a critical impediment to widespread adoption of EVs, to the best of our knowledge, this is the first attempt to propose a robotic system for automated inspection of EVBMs.
The rest of the paper is structured as follows: In Section 2 we describe our method including a systematic literature review on the use of AI and robotics for EVBM inspection, and a description of the digital twin method used in our study.The scope of the system and a description of the EVBM inspection process is specified in Section 3. In Section 4, our proposed design of the robotic work cell for battery inspection is described in detail.In Section 5, we present details of the digital twin design and simulation of our proposed automated inspection solution.Results and their research and managerial implications are discussed in Section 6.In Section 7 we discuss the study and outline its limitations.Conclusions are presented in Section 8.

Literature review
We conducted an extensive review of the literature on the use of AI and robotics to inspect EV batteries.To accomplish this, we executed the following steps: (1) retrieval of articles, (2) inclusion and exclusion of articles, (3) classification and analysis of articles, and (4) synthesis of findings.In what follows, we go through each of these steps in greater detail.

Retrieval of papers
The search for articles was carried out in three digital libraries: IEEE Xplore, Web of Science, and Science Direct.Since the scope of our study is the automated inspection of EV batteries using robots and machine vision, we searched the literature using terms that were combinations of robotics-related terms (robot, cobot), machine vision related terms (machine vision), EV battery related terms (electric vehicle, battery) and inspection.A tabulation of the search terms and article counts is presented in Table A1.

Inclusion and exclusion of papers
To keep the set of results manageable, we restricted our search to articles published in English medium journals.This yielded a total of 101 papers.18 papers were duplicates and were removed from the set of results.26 papers not related to EVs were also removed.17 of the remaining papers, not related to batteries, were removed, leaving a total of 40 papers.Of these, 19 were not related to batteries inspection.Thus, in the end, we were left with a total of 21 papers related to the inspection of EV batteries.A flow chart of this filtering process is shown in Figure A1.

Coding of papers
Next, we conducted an in-depth study of the 21 articles.Our analysis revealed six key attributes that were present in all articles: unit inspected, inspection stage, defect type, defect location, data used and algorithm used.An exhaustive list of possible values of these attributes, found in the 21 articles, is listed in Table A2.As a final step in our literature review, Table A2 was used as the coding scheme to code each of the 21 papers.In what follows, we describe this coding process and the insights resulting from an analysis of the coded papers.

Dimension: inspection unit
Figure A2 illustrates the different levels at which inspection of an EV LIB may occur.These include the electrode metal substrate, the electrode, the LIB cell, the LIB module, and the LIB pack.Each of the 21 papers were coded into one of these categories.Studies that did not belong to a specific unit were coded as 'Others.'The count of papers by unit of inspection is listed in Table A5.A look at the counts clearly illustrates that the majority of the inspection occurs for either the LIB cell or its components such as the electrode and the cell separator.Some studies examine battery pack inspection.However, inspection of LIB modules is notably absent from extant research.

Dimension: inspection stages
Figure E2 highlights the different stages in which quality inspection can occur in the process of assembling EV batteries.These stages include electrode substrate material production, electrode coating, post electrode coating, cell assembly, post cell production, LIB module assembly, post module inspection, LIB pack assembly, post LIB pack assembly, during integration of LIB pack into EV, LIB pack in EV, and LIB pack in EV in use.A count of the stages during which inspection occurs in our reviewed articles is listed in Table A4.Each of the 21 papers was coded in one of these stages.Studies that did not belong to any specific stage were coded as 'Others.'The coding results are listed in Table A4.A look at the counts clearly illustrates that the majority of inspections occur during and after cell production.Some studies examine battery pack inspection.However, no studies explore inspection during the assembly of the LIB module.

Dimension: inspection location
Depending upon the type of defect, inspection may focus on specific areas of the EV LIB such as the cell surface, the metal substrate of the electrode, the electrode, or the battery pack.The count of papers by unit of inspection is listed in Table A6.Each of the 21 papers was coded into one of these categories.Studies that did not belong to a specific unit were coded as 'Others.'The coding results are listed in Table A6.Looking at Table A6 indicates that most inspections are performed on the surface of the cell or for specific components of the cell.

Dimension: defect type
In Table A3 we tabulate the different types of defects studied in the reviewed articles.Defects are grouped under different units such as the electrode, cell, and pack.Most of the defects are related to the electrode or the cell.These are listed in Table A3 2.1.8.Dimension: data type When inspecting EV cells, values of different parameters may be measured in EV cells.These are listed in Table A7.As can be seen in the table, most of the data used are images of the batteries.A second category is data on the electrical characteristics of the battery, such as the State of Charge (SoC) and State of Health (SoH).

Literature review results
We summarise the results of our literature review in Table A8.An immediate observation is that research on the use of AI and robotics for inspection during EV battery production is scant.Secondly, most of the papers study the inspection of either the LIB cell or its components such as the electrode substrate and the cell separator.The electrode substrate is inspected for surface defects such as cracks, gaps, holes, scratches, and uneven thickness of the electrode coating (Jianqi Li et al. 2019, 2018).The cell separator is inspected for defects such as pinholes, gel particles, and contaminant particles (Huber et al. 2016).The LIB cell is also inspected for surface defects such as broken skin or weld defects in the safety vent.At the larger unit level, the LIB packs are inspected for defects such as dents, mechanical damage, and weld defects.
In summary, our literature review on LIB inspection reveals that while there is some research on the use of AI and robotics for inspecting the components of LIB cells, LIB cells, and LIB packs, there is a lack of scholarly research on the use of AI and robotics for inspecting LIB modules.A search of the literature in the popular press reveals reports of defects that creep into the EV battery during LIB module assembly (Kolodny 2018).
In December (of 2017), factory workers were manually slapping bandoliers together as fast as they possibly could, generating a lot of scrap in the process.Two current engineers told CNBC that they are concerned some of the batteries being shipped do not have the minimum gap required between lithium-ion cells.These engineers warned that this touching cells flaw could cause batteries to short out or, in worse cases, catch fire.Employees also said that quality control workers were not experienced, and two said that some batteries are leaving the factory with a potentially serious defect, a claim that An Automotive OEM vigorously denies.
-Source: CNBC (Kolodny 2018) Although the assembly and inspection processes of electric vehicle batteries remain closely guarded secrets, such media reports suggest that manual inspection of LIB modules may not be effective in weeding out defective LIB modules.This is problematic since a failure to improve the inspection process is likely to perpetuate the introduction of defects in EV batteries during module assembly.Therefore, there is an urgent need to develop solutions that help improve the battery module inspection process.
An analysis of the accuracy of AI algorithms for inspecting LIB cells, LIB cell components, and LIB packs reveals significant promise.Table A9 summarises the accuracy levels of the different algorithms that are used for automated inspection of defects in EV batteries.
Given the importance of the safety of EV batteries, it becomes imperative to replace manual LIB module inspection with automated inspection using AI and robotics.In this study, we attempt to fill this need by using digital twin methods and tools (Ding et al. 2019;Zheng, Lu, and Kiritsis 2021;Qiang Liu et al. 2019)) to design, simulate, and validate an intelligent system for inspecting EVBMs using Cobots.

Digital twin design, simulation and validation of robotic systems
A physical system's structure, components, mechanisms, static properties, and its dynamic behaviour can all be accurately mirrored by a digital twin of the system.Digital twins have two main applications in the manufacturing sector.First, digital twins are employed for digital-twinning with a real-world counterpart system (Ait-Alla et al. 2021).These systems vary in scope and include manufacturing processes (Polini and Corrado 2020), planning processes (Jingjing Li, Zhou, and Zhang 2021; Chabanet et al. 2022;Badakhshan and Ball 2022), factories (Ding et al. 2019), and supply chains (Dolgui, Ivanov, and Sokolov 2020;Kyu Tae Park, Son, and Noh 2021).Digital-twinning is used for monitoring, control, diagnostics, prediction, and cognitive augmentation (Castañé et al. 2022;Rožanec et al. 2022;Yuanfu Li et al. 2021) of these systems.In a second category of applications, digital twins are employed in the design, modelling, validation, performance analytics, and performance optimisation of future systems that do not yet exist.In this later application category, digital twins are used for product design (Tao et (Lopes et al. 2020;Maheshwari et al. 2022).This study is an example of the latter application of digital twins in which the digital twin model is created based on an idea of a future system.The steps involved in going from the concept of a system to its digital twin model include, system scope definition, concept design, digital twin system design, digital twin system validation, digital twin system simulation, simulation validation, simulation data collection, system performance analytics, and system optimisation.In our previous work, we proposed a methodology for the design of digital twins of robotic systems (Sharma 2022).In this study, we use an adaptation of this digital twin methodology for the specific case of robotic inspection of EVBMs.A detailed flowchart of this method is presented in Figure 2. In what follows, we execute each of the steps of this methodology for the design, simulation, and validation of the EVBM inspection process.

Battery inspection system scope specification
We use a systems engineering approach (Kossiakoff et al. 2011) to define the scope of the system under investigation.The system to be designed in this study is an intelligent robotic system for inspecting EVBMs.The input to this system is an assembled EVBM, and the output is a decision on whether the module is of good quality.A schematic representation of the scope is presented in Figure 3. Within the scope of the process identified in  Figure 3, the scope of this study is further limited to the final inspection step before the battery module is closed and made available for assembly in battery packs.This final inspection step is highlighted in a digital twin of a robotic system for the assembly of the LIB module (Figure 4).As is clear from the figure, this final inspection step is the only manual step in the assembly of the LIB module and combines all manual steps of assembly and inspection into a single workstation.The objective of this study is to automate this manual inspection step for a more accurate and faster inspection.

Battery inspection process specification
Having clearly defined the scope of the system for this study, we now detail the inspection process.In order to specify the inspection process, we first need a clear understanding of the battery module, its structure, and its components.For this study, we chose the TESLA Model S battery module as our use case.The top and cross-sectional views of the battery module are presented in Figures B1 and B2 respectively.444 Li-Ion cells are stacked vertically inside each Tesla model S battery module.As is clear from the cross-sectional views, the LIB cells are sandwiched between cell inserts and collector plates on both sides of the module.In addition to the individual Li-ion cells, the battery module contains a large number of additional components such as the power electronics, wiring and tubing, the thermal management system, the battery module management system, fasteners, and joining components.All these components are housed inside a box-shaped enclosure made up of Aluminum.
The cut-out image of the battery module clearly highlights that assembling the LIB module requires cramming a large number of parts and components within a small space.This raises the possibility of errors, such as loose, misaligned or missing parts, creeping in during the various steps in assembling the battery module.For example, when stacking the 444 cells that make up the module, some cells may be missed, get stacked with the wrong side up, or get stacked in the wrong pattern.Furthermore, even slight errors in stacking can result in uneven spacing between cells.At the adhesive application stage, gaps may remain, the adhesive may cover some of the cell terminals, or the adhesive thickness may not be uniform.During the wire bonding stage, some of the fuse wires may not be properly connected to the terminal or may be missing.A detailed outline of how these various defects can lead to safety incidents is outlined in the failure mode effect analysis (FMEA) diagram in Figure 5.
In the absence of an accurate inspection process, each of the defects identified in the FMEA diagram can remain undetected and cause safety incidents later on.Therefore, we closely study the FMEA diagram and list the various defects that need to be inspected for each of the components of the EVBM (Table 1).We also identify the nature of inspection required to detect each of these defects.We find that inspection tasks can be broadly categorised into visual and tactile checks.Visual checks involve looking for a defect using the sense of sight.Tactile checks require checking for defects such as the tightness of connection of a wire or a fastener.The visual checks in turn are of two broad types (1) presence/absence checks that check for the presence of a component, and (2) alignment checks that check for uniform spacing, depth, and correct patterns.

Battery inspection process description
While Table 1 lists the inspection tasks for each key component of the LIB, the order in which these parts become available for inspection depends upon their location within the assembled module and the geometry of the module.Figure B3 shows cross-sectional views of the front and back of the battery module.Since the LIB module is cuboid in shape, five of its faces (except the bottom face on which it rests) are available for inspection.Hence, a logical organising principle for the inspection process is to group the inspection tasks by each visible face of the battery module.With a view to designing the inspection process, we regroup the inspection checks in Table 1 by the top, side, front and back faces of the battery module.These checks are presented in Tables C2 and C1.In addition to listing the checks, these tables also include a description of the inspection process for each check, whether it can be automated, and whether human vision or machine vision would be superior.This assessment was made in consultation with industrial automation experts with experience ranging from 8 to 25 years in the design and implementation of automation solutions in a variety of industries.These experts included specialists in the areas of quality control, manufacturing inspection, machine vision, ML algorithms, industrial robots, and industrial cobots.This panel of experts was presented with a specification of the structure of the LIB module and its components, the FMEA diagram (Figure 5), the inspection steps (Table 1), a description of the inspection process, the process requirements and the quality metrics to be measured.
A review of Table C2 from an automation perspective reveals that most of the tasks for inspecting a LIB module are good candidates for robotic automation except for tactile tasks such as tugging and pulling to sense for tightness or snugness of fit.Furthermore, overlapping or concealed parts may not be amenable to inspection using machine vision.Having identified the inspection steps that can be automated, we next start the process of designing the robotic inspection work cell.

Battery inspection workcell conceptual design
Figure 6 presents a conceptual schematic of the robotic inspection workcell, identifying its key components.The inspection work cell is a cyber-physical system which broadly consists of hardware, software, and computer vision AI algorithms.The robotic work cell contains a robot on which an industrial camera is mounted.A conveyor indexes in battery modules one after the other.Once a module enters the robot's working space, the robot moves the camera so that it can inspect the front, top, back, and side faces of the battery module.The camera captures an image of each face and sends it to machine vision AI software for processing.The machine vision algorithms check for defects.In parallel, the human operator performs the manual checks.If the module does not pass the quality check, it is rejected and rerouted for further investigation and rework.

Robot selection
To select the robot for the inspection work cell, the inspection process requirements are translated into robot requirements that provide the lens through which different robot models are evaluated (Table 2).A careful study of Table 2 reveals that for LIB module inspection, a robot should have at least three degrees of articulation, sufficient reach to cover the dimensions of the module, precision and repeatability to take full pictures of the module faces, and sufficient payload to carry the camera.
Given these requirements, we considered articulated robots and cobots.Both robots and cobots have advantages and trade-offs associated with them (You et al. 2022).The fundamental difference is that unlike a cobot, robots cannot be in the same working space as a human for safety reasons.For this reason, robots have to be placed inside safety enclosures.From a process plan perspective, this requires separating out robotic and human activities.From a system design perspective, this requires the inclusion of transport mechanisms between work cells.Depending on the use case, sometimes these safety enclosures and transport mechanisms can cost as much or more than the robot itself.Furthermore, the need for separate work cells takes up valuable shop floor space.This separation between humans and robots is clearly illustrated in a digital twin of the LIB module assembly process presented in Figure 4.This separation is also detailed in a schematic comparison of the inspection process in a robotic, cobotic and manual inspection work cell (Figure C1).
A cobot on the other hand, allows humans to be in the same workspace, obviating the above requirements and their associated costs.Furthermore, since we do not need a safety enclosure, they can be mounted on mobile stands that can be wheeled around on the shop floor offering opportunities for a much more flexible and reconfigurable shop floor.However, for the safety of the humans around which cobots operate, they are required to move at a slower speed, which increases the cycle time of their operation.
After consulting with industrial robot experts, we narrowed our choices down to the Fanuc M-10iD/10L robot and the Fanuc CRX-10iA/L cobot (Figure E3).In consultation with these experts, we evaluated the two robots, and a detailed comparison of the features of the two robots is presented in Table 3.Based on this comparison, we selected the cobotic inspection work cell using the CRX-10iA collaborative robot.Next, in Section 5.1 we use our cobot of choice to develop a digital twin of the cobotic inspection work cell.

Digital twin design of battery inspection system
Because we use Fanuc's robot, we use Fanuc's Roboguide (RBG) digital twin design and simulation software to create our digital twin.RBG allows for the design and generation of three-dimensional models of robotic work cells using included libraries with built-in models of all Fanuc robots, generic models of robot end-of-arm tooling (e.g.: vacumn gripper and mechanical gripper),  Outfeed conveying mechanism and generic models of nonrobotic components (e.g.conveyors, and tables).Components that are not available within RBG are created in external CAD environments and imported into RBG.For instance, in our example the battery module and the HMI screen were designed in Solidworks and imported into RBG.
To create a digital twin of the inspection work cell, we first create a virtual robot world in RBG.Next, we imported a digital twin of the CRX cobot available in the RBG library.The digital twin of the robot emulates its exact dynamics, which ensures that all movements are feasible and realistic.We also imported from the library, digital twin models of the robot base, conveyors, and the industrial camera.We also used Solidworks CAD software to develop 3D models of non-standard components including the EVBM, and the HMI display monitor.These 3D models were then imported from Solidworks into RBG.
As a next step, the digital twins of the robot model and other components were used to construct a layout of the inspection workcell in RBG. Figure 7 presents this digital twin of the cobotic inspection workcell in RBG.The front For long term sustained operation, to maintain reliability, Motor Duty (overheat) and reducer life (Gear wear and tear) to be managed.Robotic simulation software, such as Roboguide can be used to determine the robot speed and rest periods to maintain reliable performance over the life time of the robot The camera payload is light and duty cycle is not a concern for both the cobot and robot.

Cost
Higher -Cobots are more expensive than robots due to the additional sensor technologies Low -This is one of the more affordable Fanuc robots.
Footprint Small Due to no safety enclosures.Large Regular robots have a larger footprint due to the safety enclosure and inter-cell transport mechanisms.

Mobility
High Because Cobots do not need a safety enclosure, they can be mounted and wheeled on carts Low This is a traditional robot that needs to stay inside a safety enclosure.This reduces its mobility Non robotic system components Cobots do not need non-robotic system components like safety enclosures and inter-cell transport mechanisms Robots need safety enclosures and transport mechanisms to convey parts in an out of these enclosures adding significantly to the total cost of the system.and top view of a digital twin of the cobotic inspection work cell created in RBG is presented in Figures 8 and 9, respectively.

Robot workcell programming
The inspection work cell is a cyberphysical system consisting of hardware and software components.We import Fanuc's R30iB Mini Plus controller to control the workings of the cobot and the entire workcell.The cobot controller is a computer that controls the cobot's motion and its behaviours.It is worth noting that a digital twin of the cobot controller contains an exact copy of the controller software available inside an actual robot controller.This enables a near-identical duplication of the motion between the digital twin and the real cobot.Furthermore, the cobot controller can work in coordination with the work cell controller, as well as controllers of other robots.Like most other robotic simulation software, RBG also provides an offline robot programming environment.We use RBG's offline programming environment to programme the controller to control the robot, as well as for workcell behaviour.Figure 10 shows a snippet of the robotic programming code written in RBG.Non-robotic aspects of the workcell were also programmed and animated to interact with the cobot, to visually convey the full operation of the workcell.Some of these nonrobotic aspects that were programmed include conveyor motion, camera activation to capture battery image, and routing of battery module based on the results of automated inspection.The cobot's movements were also adjusted  to avoid colliding with workpieces, fixtures, and other objects in the work cell.Finally, the cobot's position within the work cell, as well as its speed of motion, was adjusted to reflect a realistic cycle that takes into account long-term motor performance and the durability of the cobot gears.A key aspect of the inspection workcell is the routing of the battery modules according to the result of the inspection checks outlined in Tables C2 and  C1.To achieve this, the controller was programmed to route the module based on the results of the inspection process, as outlined in Figure 11.A key component of this programme logic is the machine vision algorithm for detecting the defects, which are described next.

Machine vision based defect detection
To detect defects, we use Fanuc's inbuilt i RVision module, which allows inspection processes to be set up using a wide range of computer vision functionality.Figure 12 lists some of the defect detection inspection processes that were setup in the Fanuc iRVision system.In what follows, we describe the setup of inspection processes for detecting defects in fuse wires.

Inspection process setup
An iRvision inspection process is made up of a sequence of inspection steps.Figure D1 shows the steps that make up the inspection process to check fuse wire defects.Snap Tool The Snap tool is always the first step in an inspection process and it captures an image of the object in the camera's view range.

GPM Locator Tool
The GPM locator tool processes the snapped image to extract the objects to be inspected (in this case the individual cell images), and their locations.

AI Air Proofing
The cells found by the GPM locator tool are then fed into the AI error proofing tool, which classifies each cell into good or bad parts.This classification is based on the training of the AI error proofing tool using images of good and bad parts.Figure D2 shows the teaching of the AI error proof tool for cells with good and defective fuse wires.

Digital twin simulation of cobotic battery inspection system
After the layout of the work cell has been detailed in RBG, the cobot programmes have been written in the offline programming environment, and the machine vision inspection process has been taught in iRVision, we run the simulation in RBG.RBG not only emulates the mechanical behaviour of the cobot, the conveying mechanism and other parts in the workcell, it also replicates the execution of the cobot programmes running on the cobot controller and the work cell programmes running on the workcell controller.In addition, the operation of machine vision algorithms is also replicated as would be exeucted by a real system.
Running the simulation allows us to detect collisions between cobots and other objects, spot areas that are beyond the reach of the cobot, simulate the payload capacity of the cobot, simulate the dynamics of the cobot, simulate the execution of the software programmes, and simulate the coordination of the cobot with the conveying mechanism and workpiece movements.In addition, the simulation allows us to simulate the execution of the machine vision programmes in iRVision.We continue to iterate and refine the work cell layout and cobot programmes till we are satisfied with the work cell operation, at which point we finalise the design of the inspection work-cell.

Results
As a final step, the simulation programme was then used to execute the finalised work cell design in a mode that captures cobot cycle time data.Because these simulated work cells are digital twins of their physical counterparts, the computed cycle times are representative of what would be obtained by setting up and running an actual physical work cell.The cycle time data are proprietary information that cannot be shared.However, a video of the inspection workcell simulation is available for viewing in an online appendix.
We now present the results of the machine vision inspection process for some of the defects.

Cell count results
In Figure D4, 241 cells have been found in a half-image of the module.A similar count of the module is done in a shot of the other half of the module.The outcome of this check is a count of the total number of cells present in the module.If there are missing cells, the GPM locator tool highlights the position of the missing cells.This machine vision inspection takes less than half a second.When compared to a human checking for 444 individual cells, it is therefore much faster and more accurate.Humans are more likely to make errors in inspecting such a large number of parts due to cognitive limitations and faltering attention and exhaustion when performing repetitive work.Fuse wire check results In Figure D3, the machine vision software highlights the cells that have a missing or defective fuse wire.The good cells are outlined by blue boxes, while the defective cells are outlined by orange coloured boxes.The software accurately identified the eight cells with missing fuse wires and one cell that is missing.Terminal check results Similar to the check for fuse wires, the check for terminals also successfully detects any defects as shown in Figure D5.The results of additional machine vision checks are presented in Appendix B.

Research and managerial implications
The proposed automated battery inspection solution allows us to achieve a fast and accurate inspection of the battery cells.Because a digital twin of the robotic workcell was used, the configuration of the battery inspection workstation, as designed and validated, is ready for immediate implementation.In our digital twin simulations, the cobot was able to complete the individual inspection tasks in sub-second time intervals with very high accuracy.Additionally, the use of a cobot instead of a robot in our solution allows human operators and the cobot to simultaneously check the EV battery for defects, further reducing the inspection cycle time.Thus, the use of cobotic solutions for automated battery cell inspection presents significant promise in speeding up the EV battery inspection process and has significant implications for scaling up battery production.In addition, the inspection process is critical to improving battery quality and safety.Because concerns about battery safety continue to be a barrier to consumer acceptance of EVs, our proposed solution also has significant implications for the widespread adoption of electric vehicles.
A review of the robotic literature and interviews with industry experts revealed that current robotic technology may not be suitable for tasks that require the dexterity of the human hand, which has not yet been effectively replicated by robots.An example is checking the snugness of fit of a bolt or a wire.Due to this partial automation of the inspection process, scaling it up for large-scale operations remains dependent on the availability of skilled human operators.However, as robotic manipulators become more like human hands, it may be possible to further automate additional steps.Additionally, the design of the battery module from a 'design for automated inspection' lens can also make some of the inspection steps more amenable to automation.A significant managerial implication of fully automating the inspection process is that scaling it up will not require consideration of the availability of skilled human operators.This presents the need to conduct research in the robotics domain to investigate opportunities for further automation of the EV battery inspection process.Furthermore, research on the design for automated inspection can help develop batteries that are amenable to automated inspection.
Since cobots can work in the vicinity of human beings, no safety fences are required, which avoids breaking the shop floor into disjointed spaces connected with interworkstation conveyance mechanisms.These mechanisms can often cost much more than the cost of the robots themselves.Therefore, even though cobots are more expensive than robots of equivalent capacity, when considering the total cost of the system, cobotic systems can be significantly cheaper.Finally, because safety fences and inter-workstation conveyance mechanisms are not required when using cobots, it is possible to radically rethink the organisation of the factory floor, resulting in significant savings of precious factory floor space.This opens up important research questions regarding factory layout design when using cobots.

Discussion and limitations
A limitation of conducting research on automating EV battery inspection is the unwillingness of firms to share their operational details.We got around this limitation by modelling our study on the TESLA model S battery whose patents are publicly available.A second limitation of conducting research on robotic automation systems is the high cost of setting up a physical prototype.In addition to being expensive, evaluating robotic automation solutions by setting up physical prototype work-cells can be time consuming, as each design tweak requires significant time and expertise.We overcame this second limitation by using the digital twin simulation method, which allowed us to design, simulate, and validate the workings of the cobotic inspection workcell.Our alternative method of digital twin simulation allowed us to design, programme, and simulate a robotic work-cell in 3-D, taking into consideration real-world requirements and constraints, without the need and expense of setting up a physical prototype work-cell.This allowed us to simulate the exact workings of the inspection solution, providing insights into its performance, enabling fast iterations and saving time.
Our study found that battery modules are currently being inspected manually.While manual inspection is an option widely used in the industry to detect battery defects, it can be time-consuming, tedious, and errorprone.Furthermore, to the best of our knowledge, there are no studies that provide a comprehensive overview of the use of AI and robotics for EV battery inspection.In this study, we have attempted to fill this gap and found that EV battery inspection is highly amenable to the use of AI and robotics.Especially, visual checks looking for presence or absence of parts, misalignment of parts, presence of cracks, scratches, or impurities, any irregularities in shape or form, are ideal candidates to be checked using this solution.
Another limitation of the proposed solution is that tactile checks that require a sense of touch, texture, or pressure have not been automated.Some examples of such checks include checking for the tightness of fit of a wire by pulling on it or checking for the snugness of fit of a part.Such checks still require a human operator and currently are not good candidates for automation.Furthermore, parts that remain concealed by other parts are not good candidates for being checked by the robotic solution.
Finally, the scope of this study was limited to the inspection of battery modules.Battery modules are in turn assembled to make battery packs which must be inspected as well.Keeping the battery pack inspection process manual will likely make it the bottleneck step that prevents a reduction in the overall system cycle time.A variant of our proposed cobotic inspection is equally applicable for the inspection of battery packs.As part of our continuing research programme on the use of robots for EV batteries, we are currently working on designing a cobotic inspection systems for EV battery packs.

Conclusion
As EV usage has increased, there has been an increased focus in the media on battery-related safety incidents such as fires and explosions.Such incidents of battery failures, without obvious external triggers, are likely to loom large in the public imagination and become a roadblock to the wider adoption of EVs.This can handicap global initiatives to slow down climate change, such as the EV-only future pledged to by a handful of countries, cities, and companies at the COP26 climate summit held in Glasgow in 2021 (James and Jessop 2021).Therefore, it is imperative for the automotive industry to produce defect-free EV batteries to ensure safe operation and wide-spread adoption of EVs.While manual inspection is an option widely used in the industry to detect battery defects, it can be time-consuming, tedious, and error-prone.In this paper, we have presented a firstof-its-kind solution for inspecting EV modules using Cobots and humans.Our study found that EVBM inspection is highly amenable to automation with the exception of some manual inspection steps.AI error proofing techniques were used, which accurately detected defects in a fraction of the time needed by human operators.We used digital twin methods to design, simulate, and validate the design of the cobotic inspection work cell.Because digital twin methods were used, the battery inspection workcell, as designed, is ready for immediate implementation in the real world.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Notes on contributor
Dr. Ajit Sharma received his Ph.D. degree in Technology and Operations from the University of Michigan, Ann Arbor, MI, USA.He also holds a B. Tech. in manufacturing engineering, a masters in industrial management, and an MBA from the Ross school of business.Before entering academia, he worked in industry for 12 years, primarily in the robotics industry.He has worked for firms, such as FANUC Robotics America, Xerox, and General Motors.He has also done startups in the areas of robotics and technology consulting.He started his academic career at Carnegie Mellon University as a professor of business technologies, teaching courses in business technology consulting and information systems, for which he received the Dean's Teaching Award.He is currently an Assistant Professor of technology, innovation, and entrepreneurship at Wayne State University, Detroit, MI, USA.His research interests are in the applications of AI and robotics for amplifying the potential of individuals and organisations.Dr. Sharma is passionate about the societal implications of technology.Since 2015, he has helped found and run LIME Lab L3C, a low profit organisation that offers pro-bono robotics and technology training to K12 kids in Detroit.

RQ 1 :
What is known about the use of AI and robotics to improve the quality of electric vehicle batteries?RQ 2: What are the gaps in knowledge about the use of AI and robotics in improving the quality of electric vehicle batteries?RQ 3: How can AI and robotics help improve the quality of electric vehicle batteries?

Figure 2 .
Figure 2. Digital twin system design and simulation method for automated inspection of electric vehicle battery modules.

Figure 4 .
Figure 4. Digital twin of EV battery module assembly line.

Figure 5 .
Figure 5. EV battery failure modes resulting from defects induced during LIB module assembly.

Figure 6 .
Figure 6.Conceptual model of a robotic workcell for EV battery module inspection.

Table 2 .
EV battery module inspection steps, process, and work cell requirements.