Markt&Technik: Herr Richter, Mr. Richter, what do the terms deep learning, machine learning and artificial intelligence mean for you personally in the context of assembly inspection, and where do you see the most important differences?
Johannes Richter, Senior Engineer in the field of industrial image processing/inspection solutions at Göpel Electronic: In the field of assembly inspection, we use the same common definitions for these three topics as in other areas of research. Artificial Intelligence is the generic and it includes the other two terms. It is the science of mechanizing the thought process. Programmable logic machines, in other words computers, have always been pioneers in this field, so it is not surprising that Artificial Intelligence is one of the core competences of computer science. Machine learning is one of the sub-areas of AI. It deals with the automated acquisition of knowledge which leads to learning. Knowledge is accumulated through observation or reinforcement learning. The most common of these are artificial neural networks (ANN), which implicitly store acquired knowledge in the weight matrices of their layers. However, numerous other approaches are also available, most of which are, unfortunately, unknown to the general public. How Deep Learning distinguishes itself from Machine-Learning or not is still a matter of debate among experts.
What opportunities do these technologies open up in assembly inspection?
AI will offer us numerous new possibilities. First and foremost, we will try to increase error detection. Intelligent systems will make it possible to improve slip-rate detection at the classification station and to fully utilize the capacities of existing inspection systems. Operators of systems will be able to do many things more easily by using intelligent assistance functions. Regular and time-consuming tasks such as the creation of test programs can be simplified. Even things such as machine and line monitoring help to ensure that production processes run smoothly.
Where is AI currently being used?
We offer test and inspection solutions for a wide range of applications and AI is represented in every field. For example, there are acoustic test systems for mechatronic components in automobiles, which are installed in tailgates, car seats, etc. Body and airborne noise measurements are carried out to check assembly quality in the production environment. Objective analysis of products by given mathematical algorithms is extended by a subjective evaluation using psychoacoustic measuring methods and Artificial Intelligence. Noise disturbances in the manufacturing process do not have a significant influence on this acoustic analysis thanks to AI. For some time now we have also been running our "MagicClick" software. This is a sophisticated AI tool for the automatic creation of an AOI test program. This automated test program generation saves time and money during the production process. In addition, we see great potential in the area of classification functions, supported by AI. Of course, we are not giving away any secrets when we say that we are working on AI applications for quality assurance in in many other areas than the electronics and automotive industries.
Which are the challenges and where are the limits?
Many people see in AI a bringer of salvation for all. But AI cannot solve every problem. However, clear benefits must be recognizable. AI must prove that it is not just hype. Artificial intelligence can only be successful if it delivers clear added value to the end user.
What risks lie in the development towards Deep or Machine Learning and AI?
Once an AI system is mature, the greatest danger is probably blind trust. Decisions left to an AI system are not necessarily objective because the basis of that decision is always a good database. If a bad solder joint is understood to be suitable for the system, it will make its future decisions on this basis. Humans must ensure that distortions of the data basis do not occur. This also means that the AI process must always be comprehensible to human beings, who must have the possibility to intervene at any time. Of course, these and other risks also lead to a certain degree of skepticism on the other side. Not least because of science fiction, many people feel exposed to an insurmountable danger. The following therefore should apply above all: Transparency and communication must be ascribed a high priority in the development of all AI systems.
What place do these technologies have on your technology roadmap?
We have already demonstrated the usability of this trendsetting technology at various points and will continue to help shape it in the future. We regard Deep Learning as one of the key technologies for our future developments in optical testing and will continue to be active in this field. But we will continue to work actively in other areas of quality assurance and also work with research institutes on developing technologies.
What does the future hold for these technologies - in general and in assembly inspection in particular?
These technologies will find applications in numerous areas of component inspection. We see a bright future especially in the field of assistance systems. In the long term, we would expect to see a development towards fully autonomous process monitoring and control. Interfaces between the individual elements of a production line will be homogenized and the boundaries between the systems disappear.
The interview was conducted by Nicole Wörner, Markt&Technik.