Artificial intelligence in quality inspection and assurance

Artificial intelligence in quality inspection and assurance

The advent of Artificial Intelligence (AI) has led manufacturers to move further with extra potential in quality inspection by reducing the human effort, time, cost and errors.

In earlier days, quality inspection and assurance in manufacturing of automotive parts, power tools, security systems, home appliances, electronics products were done by human labourers. Manufacturers required extra labour force in dealing with quality inspection and assurance which comply with the standards set by authorities and the company. It was time consuming to check each and every component in production for defects before the final product was made and the rate of human errors was more in numbers. The advent of Artificial Intelligence (AI) has led manufacturers to move further with extra potential in quality inspection by reducing the human effort, time, cost and errors. Main competitors like Bosch, Siemens, Valeo, Honeywell have taken AI into the next level incorporating into their manufacturing process, but few companies like Bosch and Siemens have taken AI into product quality assurance. The other competitors and manufactures have collaborated with different enterprises in research and development of AI with respect to manufacturing.

Industry Background and Strategic Goals

The essence of Industry 4.0 includes Industrial Internet of Things, Cloud Computing, Cognitive computing, Artificial intelligence and cyber-physical systems is the new market trend which everyone is behind. The fourth Industrial Revolution plays a vital role in determining the next big giant in the manufacturing field which has led companies like Bosch, Siemens to invest billions for the Research and Development of AI. In CES 2020, Bosch announced training programs that will make 20,000 associates ready for the AI as well as investing 100 million euros in construction of a new AI campus in Germany and also set a goal that people should trust their AI-based products. Bosch's competitors include Siemens, Delphi, Honeywell and a few others are also moving their strategies towards AI. But I find the main competitor for Bosch with regards to the use of AI is Siemes. They have also built up their own controller called SIMATIC S7-1500 to process neural networks in which the trained AI will be integrated into quality control (QC) process. The value of AI in the manufacturing market goes to worth around 27 Billion by 2027 as reported by Meticulous Research.

The current use of AI has flourished in different areas like production, QC, manufacturing process optimisation, predictive maintenance, supply chain management by the use of Neural Networks, Machine Learning, Deep Learning (DL), Reinforcement Learning and much more. As the production process becomes complex, the errors are not often detected by humans in mass production. Companies find it difficult to manufacture products economically and ecologically leading to the approach of making even the quality inspection automated. For this, the individual parts of a product are inspected in an optical procedure by processing the images against the trained AI. The use of neural networks plays a vital role in this process as it learns itself in the long run. Thus the detection of defects in components can be fixed before the final assembly of the product.

Approach, Technique and Requirements

The use of AI in quality inspection and quality assurance sum up 20.5%, which is the second-largest industrial usage of AI. Even though there are various methods to do AI-based quality inspection, the most used method today is automated optical inspection (AOI) or automated visual inspection (AVI). In general AOI uses the power of deep learning to provide automation that is faster, cheaper, and more superior. The objective of AOI is to discover most of the hidden defects during production on a day-to-day basis with the least escape rate. Even AOI is used on internal and external equipment in production lines and on finished products for quality assurance. For example in detecting missing screws, scratches, cracks, dirt, dent, defects in molding, wafer peeling, wrong labels, broken/defective products, foreign objects, and the list goes on. Business goals of these companies are to increase the production speed by reducing the time, cost, and effort spent on quality assurance and reduce the wastage in the production line. The main strategy of these companies is to produce better products with high levels of standards by reducing the cost and effort in production. This also includes a step towards reducing carbon emission from their factories with smart industrial revolution and all these sum up to add a good customer base for the company.

Deep learning and machine vision are the two technologies used in quality inspection with minimal physical equipment to automate the process. This process typically involves capturing images of the actual components without defects in different angles, resolution, dimension then pre processing these images, feature extraction, classification of images according to their type, properties etc. DL uses neural networks with different layers to mimic the human level intelligence to distinguish defects and anomalies in complex patterns with the support of high processing computer systems. DL enables machines to learn by examples as humans do normally with the help of multi-layered neural networks that mimic neurons in the human brain. This approach gives the devices the ability to recognise images and make decisions on the quality and defects in the components. In the long run deep neural networks continuously evolve as new data and images are fed into the systems by which performance and efficiency is increased. Machine vision provides image based quality inspection using computing technology to see the differences in the components moving in the production line. The set of tools include a camera, softwares and highly capable processing systems and output devices. The machine vision helps inspect objects which are too small to be noticed by the human eye with utmost care, speed and accuracy in a repeated working environment.

Deep learning has taken over the traditional machine vision based AOI as Machine vision systems fail to assess the variation and deviation between similar images due to scaling, rotation, and pose distortion on complex surfaces. This technology identifies defects in objects like scratches, dents as well and characters from the images in a complex nature. Depending on the product, DL can also be used in a combination of different tasks like object detection, barcode or serial number scanning, semantic segmentation and image classification. Bosch have developed a visual inspection assistant (ViPAS) which is the first step towards a redefined AOI-system which is task-flexible and convertible than the traditional product specific AOI systems. The objective of using these systems in product QC is to fulfill the highest quality demands of their customers. The system uses cameras that capture images of the parts to be assessed and a task specific software used for automatically detecting the defects and issues in the products. It has already shown a success rate of 99.9% after 12000 test procedures in a pilot run. ViPas is a product-flexible AOI system which can train defect detectors providing algorithms and frameworks to the users. It uses Deep Learning approaches and provides a trainable defect detection system for a new task by presenting sample data to the system. After that the ViPAS starts to learn on its own and quickly decides if a part is defective or not, eliminating the need for programming expertise or a task specific programming.

Siemens claims that QC has a lot of benefits for their customers, they use neural networks to distinguish error patterns during operation. This approach helps them to detect defects in components before the final assembly of the product. For this purpose, the individual parts are inspected in an automated optical procedure. first, adding data as images of the defective components from all angles in good quality. Second, they label these images to tell AI what exactly is in the picture, for example: a missing screw, nut or a dent in the component. Third step: let neural networks learn from the added information, when more images labelled with errors are fed into the system, the better the neural network can be trained. Final step involves integrating the trained AI into the QC process, by deploying the neural networks into the TM NPU module (Simatic S7-1500) for processing neural networks. Thus defective parts are optically detected by cameras once it goes through the production line and immediately detected and will be sorted out from the process.

Barriers and Challenges

When it comes to the ethical aspect of using AI, Bosch claims that, responsibility in this digital world is trust and trust will be important for their digital business as product quality is for their traditional business. Building product quality comes with a great challenge to the manufacturing industries. The challenges faced by these industries can be categorised into different levels, at first this approach eliminates the previous employers who were already doing these jobs manually. According to Glassdoor an average base pay drawn annually by a quality inspector sums around $ 37,961, it's actually a cost reduction for the management but to transfer the employee to a new department or remove them from the job is challenging. Second, there are several areas where AOI is considered to be of higher risk and challenge, such as nuclear power plants, nuclear weapons, aircraft and shipping parts manufacturing. Because a small miss interpretation in the system, a fault in the initial input data or development of such system can cause an injury, fatality within the factory or outside after delivering the products, even loss of expensive equipment or loss of a good customer base and reputation of the company can be the outcome. Another challenge faced by these companies is to develop and train these systems and the employees who manage them. Initially it requires a vast amount of research and development and investment of millions of dollars to develop such systems, sometimes it's hard for few industries to build such systems and compete in the fourth industrial revolution.

Realised Benefits

The benefits achieved by Bosch after implementing AI into their manufacturing process impacted in a positive way in which they improved their quality inspection performance, reduced manual visual inspection effort by the employees, improved cycle time and automated defect type detection. These initiatives impact in reducing the carbon emission from the factories as a process of smart manufacturing, which helps in protecting our environment from harmful substances. Bosch has claimed that they have reduced CO2 emissions by more than 10 percent over the last two years from their manufacturing plants, 45% test time reduction by removal of redundant test cases with a saving of $1.3 million. They have also acquired a 0% escape rate of defected components and products from their production line and less than 0.5% false alarm rate. Typical error rate from a visual inspection by humans ranges from 20% to 30%, the reason for the error can be due visually missing a product in the line, optical illusion of the human eye, imprecision of eyesight and much more. With the automated visual inspection they have reduced this error rate and the new ViPAS has achieved a 99.9% success rate in the quality assurance process. On the other hand Siemens were doing quality inspection of complex products manually, but after moving into AI based AOI they claim that their inspections are automated now and became faster, more cost effective and more reliable.


  • Bosch Media Service. (2020, January 1). CES 2020: Bosch raises the bar when it comes to artificial intelligence.
  • GlobeNewswire. (2020, March, 30). Artificial Intelligence in Manufacturing Market Worth $27 Billion by 2027 - Exclusive Report by Meticulous Research
  • Ingenuity by Siemens. (2019, November 21). Future of Automation: Artificial Intelligence is the Next Digital Horizon
  • Nanonets. (2019). Everything you need to know about Visual Inspection with AI.
  • Bosch Research. ViPAS AI in automated optical inspection.
  • Bosch ConnectedWorld.(2019). [PowerPoint slides]. Industrial AI at Bosch.
  • Bosch Media Service. (2020, February 19). AI code of ethics: Bosch sets company guidelines for the use of artificial intelligence.
  • Bosch Media Service. (2020, January 6). Beneficial AI building trust together in the digital world.
  • Siemens. (2019, November 30). Quality assurance with artificial intelligence [Video]. YouTube.