The globalization of markets does not represent only and exclusively the intensification of competition between organizations and companies. It is also great to take advantage of opportunities that certain countries offer for operations and production units to set up in their territories. It means that, for example, the same organization produces the components of its products in one country and sells or assembles the final product in another [1-4]. It happened in the organization that is the target of this study. There, the dismantled motorcycle packaging sector has the characteristics of a warehouse, with completely dismantled motorcycles. These parts form kits, which are sent from the Manaus Industrial Complex to other units in South American countries. In these other units, the kits are used to assemble the motorcycles and ship them to client organizations or end customers.

At the Brazilian unit, located in the Industrial Pole of Manaus (PIM), a city located in the heart of the Brazilian Amazon, there were problems in the manufacture of these kits due to the failure in the formatting performed by employees, often wrongly sending the quantities ordered and mixing of parts, among others. The vast similarity of identical components and the fatigue of employees contribute to these failures. It is compounded by the fact that it involves counting many parts and formatting by models, activities that require a lot of attention and concentration. It is necessary to synchronize the time of the sub-processes to finish the macro process. For this, it is essential to combine external factors since many parts come from external suppliers, so deliveries are often made after the kits are dispatched to other countries. The result is that you must pack and format the kits in a hurry to respect the transport logistics to the end customer.

Thus, this study aimed to present a proposal to control the stock of parts that leave the PIM for South American countries. The idea was to implement the RAMI 4.0 architecture, using computer vision to help solve this problem that causes a lot of customer dissatisfaction. Computer vision produces numerical or symbolic information from high-resolution images and data [5-8], configuring itself in the superior technology to solve the problems that the company faces in this issue of stock control failures.

The computer vision in the processing activity brings ergonomic gains to the employees, reducing fatigue and occupational diseases and increasing the profit margin since it will not be necessary to send parts back since the end customer is located in another country. The objectives are to reduce the transport cost and over time, the final product is successfully delivered on the stipulated date and time, so the customer feels satisfied. As computer vision surpasses the human in agility and precision, in the industry, it is applied in quality control, for counting, identification of problems or failures and in other tasks that this technology can optimize. Computer vision uses this tool to continuously supply its neural network with examples that serve as a source to interpret, catalog, classify and recognize what is in an image [9-12].

Inventory Control and Computer Vision

The literature review shows that inventory control can be a process [13,14]. The idea of ​​a process is a chained series of activities that need to be performed so that, at the end, when the last one is finished, a product is generated. Inventory control is a process because it requires a series of steps to be carried out so that, in the end, a result is obtained, the inspection of materials, as in the study by Sivamathi and Vijayarani [13] and maintenance of these materials along the supply chain, as presented in the study by Pal [14]. Thus, logical sequencing of activities and product generation are the most evident characteristics of inventory control.

The most widely known mental and physical image of a system is that of a set of things that are transformed into a certain rationally planned final product and that, if the generated result does not conform to the predicted result, presents a rectification scheme. Inventory control can also be considered a system [15]. Inventory control is a system because the desired result, which is the monitoring of stock levels, is obtained through the use of various resources, such as cameras, barcodes and QR code readers, whose data and information are used for decision-making about the maintenance, decrease or increase of stock levels. If the system fails in this monitoring, new resources and transformation processes are applied to achieve the intended result, as shown in Table 1.

Another concept of inventory control found in the literature was that of policies and procedures [16]. Policies are statements in the form of commitments that organizations and institutions make to increase the degree of trust and accreditation on the part of their operating environment about their missions. Procedures are rationalized sequences of activities that culminate in generating some intended result. Inventory control is a policy because specific commitments are signed internally or externally, such as not buying materials from slave labor or equivalent. It is a procedure because its activities are established so that there is optimal use of storage spaces and the possibility of loss and damage to materials is reduced as much as possible.

Inventory control is also a set of methods and tools [17]. A group is a collection of elements with something in common. Methods are sequentially logical procedures with established rules for the execution of each of its steps. On the other hand, tools are all kinds of physical and extraphysical (virtual) instruments used to do something. Thus, inventory control is a series of logical procedures used to achieve predetermined goals. It uses a series of tools, such as word processor tools, to produce letters, orders, memos and electronic spreadsheets to monitor and evaluate stock levels.

Several studies have described the application of computer vision and other artificial intelligence technologies for inventory and materials control. Xavier et al. [18] presented a solution for quantifying products in an industrial production line. This technology was composed of two main modules, one with the hardware infrastructure and the other with the hardware solution. The hardware solution comprised image capture and product recognition. The results showed 99.99% accuracy both in counting and classifying the products. The study concludes that the answer was much more efficient than the previous manual-based practice.

Bui and Nguyen [19] developed a study on machine learning using learning reinforcement strategies. Every time the machine hit, a backup was directed at it. In this way, the algorithm optimized the results by continually correcting itself so that it could reach the best possible solutions. Among them was inventory management, which fed car trading strategies that resulted in real-time online price optimization. They considered products with low inventory levels and variations resulting from the competition's advertising campaigns.

The study by Martynov et al. [20] developed a solution to determine the prediction of failures in materials in stock. They recorded the temperature and other parameters that characterized the stocks. The management and inventory control system was centralized and had powerful data processing centers and high-speed communication networks. It allowed the creation of an expert system capable of analyzing the technical state and predicting its resource. 

Rizzi [21] tried to describe the supply chain from a physical representation, highlighting the actors involved and the physical structures operated by them. It was all established from the use of computer vision.

The study by Carvalho et al. [22] proposed a survey method based on digital stereoscopy of images captured by uncrewed aerial vehicles to control the charcoal stock. The created method presented a high quality and recorded the material stock with forecast and geometric accuracy. The results showed that estimating the volume of coal piles can replace the conventional method of survey, with considerable gains in forecasting the volume of stock, inventory frequency and work safety.

Table 1: What is Inventory Control

Source: Data collected by the authors

The Practical Case

Computer vision can help alleviate or eliminate production processes' severe and often mundane problems. Here we will report an occurrence, with its symptoms and causes, proposing computer vision as a Rami 4.0 tool that can be applied to this and other problems.

The Problem

Factory Alfa (fictitious name) is located in the Industrial Pole of Manaus. It carries out activities to export dismantled products to South American countries. These products were packaged and dismantled, in the form of kits, to be assembled in the importing countries. It presented the company with the challenge of packaging the kits with all the components. Therefore, a lot of precision was needed so that the importing companies had no problems during the assembly process of the final products.

However, a severe problem was found in a certain period: the kits arrived with a series of non-conformities. Most of these nonconformities were in the form of duplicity and lack of components in these kits. The problem is even worse because the transport time between the Manaus plant and the South American destinations took, on average, 25 days, given that the waterway modal was used, first by the river, in the Amazon rivers and then by sea. For the factory's customers to be able to fulfill their commitments to their customers, the missing parts would have to be sent by air, much more expensive than the waterway. These non-conformities, in turn, generated undesirable consequences for Factory Alfa and its customers, mainly the increase in costs and customer dissatisfaction.

The Causes of the Problem

Analyzing the problem allowed the causes of the problem to be identified. The Ishikawa chart, also known as Fishbone, was used for this. The results showed failures in four of the six analytical dimensions of the cause-effect method: method, materials, environment and labor, as shown in the data contained in Figure 1. The failure of the technique used to control the materials was because of the manual conference and supervision schemes made with the naked eye. Measurement scales were used to measure tiny parts. In this way, the control process flowed. It was doomed to the generation of failures in the process that every limitation of the human senses presents. The weighing of the parts was added and done manually, configured in yet another way of trying to detect the failure in time. As it was an archaic procedure, losses continued to occur.

Failures arising from the Materials dimension of the cause-effect method happened because the main inventory control activities were exclusively manual. The jobs were all manual. As a result, measuring scales were manually used to gauge the measurements of the pieces. In addition, the precision scale used to help count the units of the parts to be sent was constantly out of calibration. As these scales were not calibrated promptly, all activities were compromised since the calibration service was performed by outsourced companies that were not immediately serviced. It forced the need to stock spare scales.

The failures generated by the Operations Environment were because the packaging and preparation process of the kits required a high degree of concentration from the operators. It turns out that the operations environment had many serious noises that distracted and dispersed operators' attention. In addition, insalubrity was considered very high, which bothered the workforce, especially the excessive heat in the Amazon between 11 am and 4 pm. The most noticeable symptom was the intense fatigue of the workers.

The flaws detected in the dimension of workforce analysis were mainly due to the very low professional qualification of the employees who operated in this critical industrial activity. There was a lack of knowledge and skills about the correct and proper handling of the tools, especially those related to verifying the configuration of precision balances.

The issue of the non-conformity of the kits failures in the mixture of models of different pieces and fatigue of the collaborators in their respective activities. Figure 1 shows the Ishikawa diagram with the problem and its separate flaws, organized by analytical dimension. It can be cited that some kits were packaged and missing some components. It makes it impossible to assemble the products at their destinations, creating with the final customers of the importers.

The vast similarity of almost identical parts and the fatigue of the collaborators contributed to the spread of failures, causing errors in part counts, model formatting and increased time to complete the process, among others. External factors also generated terrible consequences, leading to new failures, especially parts from other suppliers, which often arrive at the last minute and have to be packed. In these cases, the kits are hastily formatted to respect the transport logistics to the end customer. The consequence of all this was customer dissatisfaction for reasons of non-compliance. Internally, there was an increase in occupational diseases caused by employee fatigue and additional financial costs, especially with air transport and rework outside the company's production plans.

Symptoms of the Problem

Symptoms represent the different forms of manifestation of a problem. For example, fever is always a symptom of something wrong in the human organism. 

The most visible symptoms of nonconformities in the kits were customer dissatisfaction, waste of money and a lot of rework. Due to the nonconformities, the automakers could not finalize the assembly of the motorcycles; by extension, the automakers were unable to meet the orders of their customers in the agreed time, generating, also for automakers, dissatisfaction in not receiving the product and not serving their customers, four other apparent symptoms.

The waste of money was due to the factory's need to reprogram itself to reduce the unsatisfactory situation of its customers. The decision was taken to exceed its costs, extrapolating the production budget, to serve the South American customer in a way that would not delay that customer's production. As it was about logistics from one country to another, it was necessary to spend even more money on air logistics so that the missing components of the kits could arrive on time, which considerably increased the unwanted cost of the company.

Internally, the cost of rework meant that different work schedules had to be rescheduled to assemble the kits. There were, therefore, regular production crews and more great crews to produce what was not included in the dispatches. Thus, Industria Alfa had to pay its employees overtime to correct an error that could not have been noticed. Errors are generated exclusively by rudimentary human failures.

Figure 1: Causes of Stock Control Failure

Source: Prepared by the authors

Problem Solving Proposition

The solution to the problem of non-conformity in the kits could be performed through the implementation of Rami 4.0, using computer vision. The VDMA 24582 specification [23] describes an approach to developing a Fieldbus-neutral reference architecture for the condition monitoring system. The starting points for this approach are information about conditions currently available in system and field devices from component manufacturers of technologies relevant to automation. It is focused on the fundamental questions of how these components can be given a uniform structure to allow communication between them. Computer vision produces numerical or symbolic information from high-resolution images and data. It can make processes more agile and simple, replacing any visual activity. Unlike a human operator, the computer can see several things simultaneously, with a high level of detail and without getting "tired."

The problem was characterized by the fact that (a) there was a lot of similarity between common parts, (b) the kits were made up of parts of similar models, (c) the production process worked with tiny motorcycle parts and (d) it consisted of large amounts of small parts. All this represents a high probability of generating many failures in counting and separating the models to be sent.

The proposed solution is implementing industry 4.0 in the packaging process through Rami 4.0 and applying computer vision to filter the similarities of the parts by models, dimensions of the elements and countless other parameters considered fundamental for the control of the stock. As computer vision surpasses humans in quality, agility and precision, it is also applied in quality control for counting, identification of problems or failures and other tasks that can be optimized. Computer vision uses these capabilities to continually supply its neural network with examples, which serve as a source for interpreting, cataloging, classifying and recognizing what is in an image.

Results to be Achieved

The application of computer vision to solve the problem of non-conformity of kits intends to generate as main results the saving of time, improvement of quality, reduction of unexpected failures and reduction of unnecessary activities.

Time-Saving

The Time saved by implementing computer vision in the parts packaging process is remarkable. Additional time and results are invested in the continuous improvement of procedures in search of increased productivity with excellence. The intention is to increase productivity with quality, allowing the employees of each workstation to enjoy the interval rest provided in the labor laws. It is also possible to perform labor gymnastics-the work gains in quality, which generates high performance from the production team.

Quality Improvement

The most definitive proof of quality improvement is that the end customer considers himself satisfied with the products he receives. The parts arrive at their final destination with inspection by computer vision without any unforeseen events, with impeccable quality and without impact on the physical quality of the products. It also impacts the physical and mental quality of the employees involved since, with the implementation of rami 4.0, fatigue is reduced, which is very common without this technology.

Reduction of Unexpected Failures

Unexpected failures are not only reduced but are also eliminated from the production process of producing kit packaging. Thus, losses in the packaging of similar parts, difficult to detect with the naked eye, failures in counting the components that make up the kits, sending kits with missing details, production line bottlenecks, generating costs and inconvenience and end customer dissatisfaction become things of the past.

Reduction of Unnecessary Activities

Implementing Industry 4.0, starting from the Rami Architecture 4.0, reduces unnecessary activities. In addition, the exposure of employees to the unhealthy work field (as they are activities that require non-interference with ventilation). Ventilation does not help the efficiencies of the scales, for example, because they are precise. In this environment, the tiring activities require the employee to count and sometimes recount hundreds of tiny pieces to form several kits, compromising their health, quality of life and the process. These activities are all eliminated with computer vision being applied to inventory control.

This study presented a proposal for computer vision to control the stock of packaged kits for export made by an industry located in the industrial hub of Manaus. The purpose is that this solution will bring actual results saving time, improving the quality of the products, reducing unexpected failures and reducing unnecessary activities. As computer vision does not work alone, it finds support in the systematics of connections and interconnections in layers, from the product to all the links in the supply chain, as recommended by the Rami 4.0 architecture.

Acknowledgments

We greatly appreciate the financial, managerial and institutional support of our partners: Envision Indústria de Produtos Eletrônicos Ltda., Centro Internacional de Tecnologia de Software do Amazonas (CITS AMAZONAS), Fundação de Apoio ao Ensino, Pesquisa, Extensão e Interiorização do IFAM (FAEPI), Pró-Reitoria de Extensão do IFAM (PROEX-IFAM) and Polo de Inovação do IFAM.

1. Annen, K. “Social capital as a substitute for formality: Evidence from Bolivia.” European Journal of Political Economy, vol. 31, 2013, pp. 82–92. https://doi.org/10.1016/j.ejpoleco.2013.04.002.

2. Boluwatife, G.A., Oladeji, S.I. and Ogbaro, E.O. “Trends and patterns of foreign investment inflows in Nigeria.” Sapientia Global Journal of Arts, Humanities and Development Studies, vol. 5, no. 2, 2022, pp. 251–260.

3. Hilmi, A. “The alfredo namitete agroecology credit system: A new business model that supports small-scale lending.” Sustainability, vol. 11, no. 15, 2019, pp. 1–2. https://doi.org/10.3390/su11154062.

4. Bozzi, C. et al. “Fashion E-Tail and the impact of returns: Mapping processes and the consumer journey towards more sustainable practices.” Sustainability, vol. 14, no. 9, 2022, pp. 1–39. https://doi.org/10.3390/su14095328.

5. Liu, H. et al. “A Numerical measurement method for dynamic granular materials based on computer vision.” Materials, vol. 15, no. 10, 2022, pp. 1–17. https://doi.org/10.3390/ma15103554.

6. Sabour, A. et al. “Maps vision: A computer vision-based system for detecting discrepancies in map textual labels.” 2022 23rd IEEE International Conference on Mobile Data Management (MDM), Paphos, Cyprus, 6–9 June 2022. IEEE, 2022. https://doi.org/10.1109/MDM55031.2022.00065.

7. Rodrigues, F. et al. “Application of deep learning approach for the classification of buildings’ degradation state in a BIM methodology.” Applied Sciences, vol. 12, no. 15, 2022, pp. 1–18. https://doi.org/10.3390/app12157403.

8. Choi, S. et al. “Point cloud-based lane detection for optimal local path planning.” 2022 International Conference on Electronics, Information and Communication (ICEIC), Jeju, Republic of Korea, 6–9 February 2022. IEEE, 2022. https://doi.org/10.1109/ICEIC54506.2022.9748306.

9. Patange, G.S. and Pandya, A.B. “How artificial intelligence and machine learning assist in industry 4.0 for mechanical engineers.” Materials Today: Proceedings, in press, 2022, pp. 1–4. https://doi.org/10.1016/j.matpr.2022.08.201.

10. Mendes, J. et al. “AI in Breast Cancer Imaging: A Survey of Different Applications.” Journal of Imaging, vol. 8, no. 9, 2022, pp. 1–20. https://doi.org/10.3390/jimaging8090228.

11. Baytamouny, M. et al. “AI-Based home security system with face recognition.” 2022 6th International Conference on Trends in Electronics and Informatics (ICOEI), Tirunelveli, India, 28–30 April 2022. IEEE, 2022. https://doi.org/10.1109/ICOEI53556.2022.9776900.

12. Santamaría, G. et al. “Combining Image Processing Techniques, OCR and OMR for the Digitization of Musical Books.” International Workshop on Document Analysis Systems. Edited by S. Ichida, E. Barney and V. Eglin, Springer, Cham, 2022.

13. Sivamathi, C. and Vijayarani, S. “Promoting Business Activities Using Utility Mining Techniques.” Handbook of Research on Promoting Business Process Improvement through Inventory Control Techniques. Edited by N.H. Sha and M. Mittai, IGI Global, Hershey, 2018.

14. Pal, K. “Evaluation of RFID Tag Anti-Collision Algorithms in Supply Chain Automation.” The IoT and the Next Revolutions Automating the World. Edited by D. Goyal and S. Balamurugan, IGI Global, Hershey, 2019.

15. Hooshangi-Tabrizi, P. et al. “Two-stage robust optimization for perishable inventory management with order modification.” Expert Systems with Applications, vol. 193, no. 5, 2022, 116346. https://doi.org/10.1016/j.eswa.2021.116346.

16. Julienne, R. Inventory Management and Organizational Performance: A Case Study of Kabgayi Printing House. Master’s thesis, Mount Kenya University, Tika, Kenya, 2017.

17. Santos, J.O. et al. “A Importância do Gerenciamento de Estoque no Âmbito das Organizações.” Administração e Gestão Pública. Edited by R.M.S. Santos and J.O. Santos, Geades e-Books, Patos, 2017.

18. Xavier, R.J. et al. “A deep learning approach for automatic counting of bales and product boxes in industrial production lines.” International Conference on Computational Science and Its Applications: 22nd International Conference, Malaga, Spain, 4–7 July 2022. https://doi.org/10.1007/978-3-031-10522-7_42.

19. Bui, T.H. and Nguyen, V.P. “The impact of artificial intelligence and digital economy on vietnam’s legal system.” International Journal for the Semiotics of Law–Revue Internationale de Sémiotique Juridique, 2022, pp. 1–21. https://doi.org/10.1007/s11196-022-09927-0.

20. Martynov, I. et al. “Development of a procedure for determining the pre-failure condition of the axle boxes of railway rolling stock.” Communications–Scientific Letters of the University of Zilina, vol. 24, no. 1, 2022, pp. B87–B93. https://doi.org/10.26552/com.C.2022.1.B87-B93.

21. Rizzi, A. Supply Chain. Springer, Cham, 2022.

22. Carvalho, L.M.E. et al. “Charcoal heaps volume estimation based on unmanned aerial vehicles.” Southern Forests: A Journal of Forest Science, vol. 83, no. 4, 2021, pp. 303–309. https://doi.org/10.2989/20702620.2021.1997067.

23. Heidel, R. et al. Industrie 4.0: The Reference Architecture Model RAMI 4.0 and the Industrie 4.0 Component. Beuth; VDE, Berlin, 2015.