Shaping The Future Of Manufacturing With Artificial Intelligence
When I’m asked what the factory of the future will look like, I always say: smart, lean, and green. These are all characteristics that already describe the factories in our global production network at Vitesco Technologies. But when it comes to artificial intelligence (AI) in manufacturing, I think we’re still at the beginning of an exciting journey. AI systems will completely change the way we run our production and enable greater efficiency by improving processes, providing real-time insights, and even predicting the future.
That’s why we have set up competence centres in Germany, France, and China where we develop such innovative AI solutions and implement them in our production lines - or as we like to call it: We are giving our production systems a brain!
We see the greatest opportunities for us to profitably use AI in the areas of process monitoring and quality control. Here we are focusing on three key projects:
• Predictive Quality
• Prescriptive Quality
• Predictive Maintenance
Prescriptive quality is about deriving actions based on the data collected to ensure that a future desired outcome does or does not occur. The question is, how do I have to adjust my process parameters based on historical data so that I do not produce defective parts. This is also called closed-loop process control.
The first step is to automate and objectify quality decisions to eliminate subjective influences caused by these manual inspections. Because the more operators are performing manual inspection and the greater the diversity of parts to be inspected, the higher the potential error rate. With AI, we can reduce human judgement bias by 300 percent.
Subsequently, further process data must be generated from individual machines or lines to better understand the processes and their influences. Analysing these enormous amounts of data and recognising patterns using neural networks enables optimised control of processes and can reduce scrap before it is produced.
We have already started to replace manual inspection with automated visual inspection using AI. In the process, algorithms have been developed and trained based on a large number of images, which can then independently distinguish between good and bad parts.
Complex manufacturing processes and products have between 25,000 to 100,000 production variables with different characteristics that must simultaneously match up to 500 final quality parameters. Our goal here is to implement an automatic calculation of an individual quality index for each product in each process step based on all relevant product-related data with AI. For us, it is the key player in detecting anomalies in this complex environment.
It should be noted that not all anomalies are defects, but all anomalies are an indication that something unusual is happening. AI allows us to identify and analyse these correlations.
In addition to installing a suitable infrastructure for data collection and processing, the challenge is to constantly learn and recognise patterns from the ever-changing data in order to ultimately react in real-time. With the implementation of conditional anomaly detection, we expect up to a 30 percent reduction in scrap and rework while reducing raw material consumption and time to market.
Predictive maintenance is the in-depth analysis of machine data to predict future machine deterioration and/or failure to avoid unexpected shutdowns.
In the past, we have focused on predictive maintenance projects mainly on the most important bottleneck machines. But to generate a universal predictive maintenance algorithm and make valid predictions, you need data from a large set of identical machines. You will never be successful with a small number of machines to create algorithms for the entire population. As our data experts always say: big data only works with big data.
So, the first step is to consolidate the immense volumes of data in the cloud, where the current operating states can be determined and visualised in order to detect and communicate unplanned downtimes.
In a second step, we are working on the development of neural networks to determine the “state of health” of the machines from this data. That means the machine or the robot is working flawlessly or is a breakdown to be expected soon.
Our experts and data specialists are working hard to make our production systems smarter. Because ultimately, these process optimisations always go hand in hand with yield increases.
But it also requires a precise evaluation of where and in which processes the use of AI adds value. In the end, AI is not about machines learning on their own and replacing people. The cooperation of humans and machine is the decisive factor for success. To condense the flood of data in production, to work out trends, and to make the right decisions based on the given data describes for me the future of manufacturing. If you really want to gain significant benefits with AI, the organisation must develop into a learning organisation. That means machines learn from people, people learn from machines, and machines support people to make the right decisions.