Dr. Krishna Iyengar is the Co-founder and CTO at Jidoka Technologies, a Chennai-based next-generation start-up, delivering cutting-edge engineering solutions to automate the process of visual quality checks by leveraging Deep Learning, AI/ML, and Analytics. He is a technical innovator and leads the product development and data science divisions at Jidoka Technologies. He currently has two patents pending in the area of Artificial Intelligence in Machine Vision. Dr. Krishna is also an expert for manufacturing and automation and has a patent with IBM, New York, in the domain of semiconductor manufacturing.
AI in manufacturing in general as we are aware has several benefits, including, delivery of 98% or higher accuracy in the QC process, enabling 100% visual inspection, and avoiding backlogs and bottlenecks with 24X7 deployment of the solution. An increase in efficiency, throughout, and high-quality products are clearly observed with the use of this technology resulting in savings, higher revenues, and providing a competitive edge. It addresses the challenges related to human fatigue, health, and safety issues, allowing skilled manpower to take up more creative and high-end roles leading to better employee satisfaction and improved quality rather than using manpower to perform low-end, repetitive tasks. Many a time, even complicated inspection processes, a result of complex manufacturing processes can be managed with much lesser human intervention.
While we understand the advantages of adopting AI-enabled solutions for visual inspection and QC processes, manufacturers are yet hesitant to go ahead with it, and this is mostly due to a lack of information on the subject. Additionally, there are age-old belief systems that tend to make adoption of AI not a comfortable thing for many organizations, as they believe they are not ready for it. But these are all issues that can easily be resolved by busting some basic myths and educating people on what basic things they need, to be able to aid adoption
Talent shortage adds to the delay
With the field being relatively new, the dearth of talent exists as solution developers with the knowledge to implement Machine Learning and Deep Learning are less in number. The few who have the skill sets to understand the sophistication in configuring and implementing AI-based solutions on a large scale, demand very high salaries, adding further cost to the company.
Investments in training employees and the related infrastructure drain the organization’s financial resources and lead to delay in the adoption of AI tools. It is best, therefore, to rely on experts who will be able to do a turnkey job, working with your internal team to help achieve your manufacturing objectives
Lack of sufficient defect data
Manufacturers all agree to the scarcity of relevant data relating to different defects required to train AI systems. Even those who have these documented, know that it is not exhaustive. The insufficient data scenario will give rise to more false positives, where even a good product will be mistaken to be a defective one and vice-versa, leading to wastages or products already inspected and classified as good products still retaining missed defects. Other variations due to complex surfaces may not have been captured by the system, delaying the process. The ’silos’ nature of data that resides in different business units or departments is not made access to all making it difficult to streamline and interpret.
Collecting relevant data from all sources and in all formats, and curating it for training has to be exhaustive and the camera-based computer vision may not have captured all the defects. In such a situation, all possible categories of datasets representing detailed features may not be available to build stable AI models.
This can be achieved by working with experts who will not only help build an exhaustive defect document that includes the existing defects that are available, the experience of the subject matter experts as well as in-depth research using AI tools to identify what is being missed out.
Challenges in deployment
Upon comparing AI-based visual inspection to Traditional methods of automation, 100% accuracy is not achievable if the visual inspection is subjective in the former, which has an algorithm meant to think like humans. To go from current levels to 100% accuracy is possible only with a comprehensive shop floor digitization and transformation journey that balances both the Manufacturing and Economic factors of the organization. Another aspect that must be taken into account here is the existing infrastructure and how much of it can be leveraged or repurposed to minimize the cost of digitalization.
Challenge of integrating AI with the infrastructure
Not all manufacturing facilities have the appropriate infrastructure and relevant equipment to implement AI. Integrating the technology with the existing infrastructure is another area of concern. With the rapid evolution of AI technology, there is the possibility of the system becoming obsolete within no time. What organizations need is the flexibility to get upgraded or modified without them having to invest in additional equipment; a solution can repurpose the outdated AI models and scale beyond the initial deployment or pilot program.
The challenges seen in the deployment of AI-enabled technologies can be overcome with timely intervention by applying the most appropriate measures. The first step to be taken on the adoption of an AI-based model is to ensure that the employees in different business units, across the organization, understand and embrace the concept and its usage. It should be communicated as part of the mission and vision of the organization with acceptance from all, top-down, with leaders leading the transformation by putting the appropriate policies and governance in place.
While providing training in the process of developing in-house capabilities, it is also very crucial to leverage third-party resources or join hands with an external AI solutions partner. The proven AI partner in the space will have highly skilled resources, expertise, and viable solutions at scale.
The benefits of using AI-powered quality control solutions outweigh the challenges and organizations are seeing the benefits of such systems. Typical benefits are decision-making at speed, consistently and accurately that is not possible by humans. Community knowledge of defects can be captured and reused across lines and factories. Other advantages include the availability of Digital twin for insights and improvements, reduction in employee fatigue, and retraining costs in addition to an increase in yield, leading to higher revenues.