Depending on your perspective, your thoughts on artificial intelligence (AI) probably fall somewhere between the technology being an abstract threat or possibility, and a real-world solution with concrete use cases where you may not even know the technology is at work. This article summarises the case studies presented at the AI workshop at ARC’s recent virtual European Industry Forum and shows potential usage of AI in machinery applications within today’s smart factories.
One of our key findings is that a clear use case for AI is needed and the target established for that use case must be met to determine the ultimate success of the AI project. While the use case must be defined clearly, there is almost no limitation to the types of applications AI can support. When edge and cloud are leveraged in the right way and connectivity to other systems assured, the possibilities are almost endless.
Finally replacing muscles and brains
Starting around 2009, people began talking about the fourth industrial revolution, Industrial IoT, and other related concepts. However, in retrospect, the second and third industrial revolutions largely just replaced human muscle and manual labour with machines and computers that basically repeat pre-programmed behaviour. While the fourth industrial revolution increased the level of digitalisation, until recently even the most educated machines and computers did not make human-like decisions. Now, with AI entering the plant floor, we’re finally starting to use digital technology to replace not only muscles, but also brains. Most experts agree that while AI will become deeply embedded across industrial and other applications, and initial use cases have emerged, AI in manufacturing today is still a niche technology. In addition to the numerous AI-related sessions and ad hoc surveys at our recent industry forums, ARC is conducting an ongoing online survey for industry participants to identify and support the most suitable applications.
What will AI look like in the future?
When asked how they believe AI will be used in future, more than 100 industry participants shared their responses.
Most agree that machinery will have AI in the future, but there is no overall agreement whether AI will be used in most machinery or just for high-end machinery. One possible explanation for this is that people have different perspectives on what constitutes ‘high-end’ machinery. Also, we intentionally did not specify a time horizon for this question. ARC’s initial conclusion from this is that AI applications will start with more high-end machinery and then gradually migrate toward simpler machinery, such as palletisers and packaging machines.
In contrast, there is almost total agreement that AI will be deeply embedded. This may be in the controller, the engineering tool, or even embedded right into the device.
Technical constraints do not seem to be a big issue among our survey participants, but cultural issues are. ARC agrees with this. AI will take decisions away from the well-understood controller and, especially when deeply embedded, the results of the AI techniques are not 100 percent transparent. This is a real drawback in a generally conservative industry such as industrial automation.
Another finding from our online survey is that unclear use cases are among the top inhibitors for AI in manufacturing. This lines up with ARC’s observations from other industries: adopting new technology for technology’s sake will not succeed. Hence, our European Industry Forum has featured use cases and best practices from leading OEMs, end users, and suppliers of AI, which we will summarise and discuss below.
Where we are with AI in manufacturing
Many technology suppliers now offer AI-enabled products and many machine builders have started to evaluate the technology. However, there are several roadblocks; most prominent among these are the lack of data scientists, lack of available data, legal aspects, human factors, and finally, unclear use cases.
ARC market research on AI in machinery applications identifies the current distribution of AI – maintenance applications in particular are prominent in the market.
Studies from ARC European Forum
We segmented the following case studies from the ARC European Forum by application, rather than company. The expert presenters were Andreas Geiss from Siemens, Maarten Stol from BrainCreators, Prabhu Venkatra-manan from Larsen & Toubro (L&T) Construction, and Sander Aerts from Toyota Material Handling.
According to BrainCreators, over half of the quality checks in manufacturing involve visual confirmation, which are an easy target for AI. A challenge here is performing the needed quality checks with increasingly small batch sizes and higher variances in production, where a combination of expert know-how and AI support is the right solution. One of the key challenges in the industrial world is the ‘codification of knowledge.’ This means documenting the tribal knowledge of engineers and others in written or other format so it can be shared across the entire organisation. Maarten Stol referred to this as ‘turning domain expert knowledge into digital intelligence.’
L&T; Construction uses AI and image processing techniques to ensure quality in welding processes. While this can be done in a pre-defined setting for in-plant welding machines, L&T; is challenged to analyse welds in pipelines and other typically ‘one-off’ welds. While the images are taken locally at the edge, analysis is done in the cloud to be able to train and use more computing power.
One step further to testing every single product is an approach from Siemens in its Amberg factory, an electronics production site. In this high-speed production environment, testing each piece is time and cost intensive, so AI is used to predict quality using production data and existing quality test results.
ARC estimates that maintenance is the largest implementation area for AI applications in machinery applications, where it can bring substantial savings.
BrainCreators showed its solution for cabinet inspection, which allows real-time, proactive asset management based on maps of assets. Another pain point is the post-inspection workflow (the aggravating and time-intensive, but essential paperwork) needed to comply with regulations. Here, AI can help and support the maintenance staff, so they can focus on key value-adding work.
Warehouse logistics, MES, and ERP applications
One highly discussed application at the Forum was Toyota Material Handling’s forklift application. The application describes the development from manual forklift trucks to guided vehicles to autonomous vehicles. This spans several applications, including safety, collaboration, and operational optimisation. Training, verification, and optimisation are performed in the cloud using a digital twin. There is on-board intelligence and cloud-based communication, enabling swarm intelligence and distributed learning.
The cloud-based learning with a digital twin not only shortens project times, but also provides Toyota with a lifecycle service opportunity using failure analysis, preventive maintenance, and optimisation. Further integration with ERP allows fully automated logistics that also work with smaller batch sizes.
AI is increasingly popular to improve usability of systems, from MES up to ERP. In addition to pre-generated interfaces and reports, AI can help answer questions such as “How much material is used?” Contextualisation is key here and needs close collaboration between users such as L&T; and the AI experts, Mindtree, in this case.
In addition to quality control, AI can be used to fill in and generate reports to free workers from paperwork. L&T; is also using AI to read and better understand longer documents, such as tenders. Here, AI can be used to assess the risk of a given project, support the pricing process, and help quickly answer questions about tenders.
L&T; uses AI to identify workers via cameras (facial recognition) on construction sites. This can be used for workforce management, to identify workplace safety issues, or even COVID-19 protection. It can also identify workers using unsafe devices, such as mobile phones, staying too long in unsafe areas, or not wearing helmets or other safety equipment.
Energy management benefits
Reducing waste in discrete manufacturing processes is among the best ways to reduce energy consumption as scrap is reduced and no energy is wasted producing the wrong parts. In addition, monitoring and predicting quality enables a smooth and energy optimised workflow. While our forum workshop didn’t include a dedicated energy management example, many of the cases did actually reduce the energy consumption.
Operational simulation and optimisation
While consumers expect consistent product quality and taste, natural variances of ingredients impact both in food and beverage products. This is a particularly acute problem for global brands. Working with a brewery customer, Siemens developed an AI-based solution that aims to guarantee consistent quality and taste. An additional benefit from this project was to reduce loss of know-how, since the solution was written down and codified.
Siemens also cooperated with a press manufacturer to optimise machine operation. Here, AI learns the relationship between multivariate data streams and product quality and sets the press parameters accordingly. As every change in parts or material impacts product quality in a press shop, this example shows that AI can increase the throughput of smaller batch sizes while guaranteeing product quality.
In robotics, motion planning is used to find the optimum configuration of sequences needed to move an object from source to destination. The Toyota Material Handling use case illustrated how this optimisation solution can be applied to logistics in warehouses.
The ‘tip of the iceberg’ for the smart factories of the future
The use cases presented at our AI workshop at the ARC European Industry Forum span a range of industries and AI techniques. It is likely that these only represent the tip of the iceberg when it comes to potential applications for AI in machinery, and our survey results seem to support this. What the examples have in common is a clear and dedicated use case, whether to improve quality, help ensure worker safety, lower operational costs, lower scrap rate, or reduce the cost of maintenance.
The examples all show that while most of the applications do not really replace human brain power per se, they certainly support it. For the foreseeable future at least, it’s not likely that AI will supplant human problem solving and creative thinking, but will supplement both, particularly for applications with a clear and dedicated use case.
SOURCE: IT in Manufacturing