One of the main objectives of European industrial companies should be to use artificial intelligence (AI) primarily to optimize existing plants, machinery or tools. This is because companies want to keep their investments in Industry 4.0 and Industrial IoT (IIoT) manageable. Many decision-makers will therefore rightly want to optimize an inventory in the sense of a retrofit in the best way possible, instead of procuring a new one. So how can an existing component, such as a gripper, deliver new added value using AI? Or to put it another way: How does an industrial gripper become a smart industrial gripper?
Grippers usually work quite reliably, but they do not always work as precisely as would be necessary for some sensitive components. The Data Science team at Körber Digital has therefore investigated how complex gripping processes can be and how they can be carried out more intelligently: from data analysis and evaluation to the characterization of gripping processes, and the identification of additional applications.
The identification of faulty components is just one of many possibilities opened up by smart grippers. Sven Warnke, Data Scientist, summarizes the key questions: “How can faulty gripping processes be detected and avoided? How can a successful gripping process be characterized? And: which additional statements and findings can be gained?”
“The first step was to review, evaluate and interpret existing data. The basis for this was close consultation with experts and a fundamental familiarization of our Data Science team with the electromechanical processes and interrelationships of a gripper,” explains the data expert.
During the subsequent visualization of the data, clear patterns emerged. Warnke explains, “A gripping process can be divided into the three phases: ‘gripper closes,’ ‘object hold,’ and ‘gripper opens’ on the basis of the measurement data. We have developed an algorithm from this which automatically performs this subdivision for the entire data.”
Based on this, the gripping processes were analyzed in more detail: In an iterative approach, hypotheses were continuously developed, which were either substantiated or refuted by further representations and analyses.
The detection and avoidance of errors, for example, by outlier detection, is one of the basic disciplines and frequently applied methods of Data Science. “In the case of grippers, we thought from the opposite direction and defined successful gripping processes first and foremost, so that we could gain more valuable insights than with pure troubleshooting.”
To characterize successful gripping processes, Körber Digital identified two approaches that complement each other: In the first approach, the acceptable value range for measured values, such as the gripper position or the gripper current, was identified at certain times during a gripping process – resulting in a kind of benchmark for a successful gripping process. The second approach describes a successful gripping process on the basis of certain gripping process characteristics, known as “features.” The corresponding value ranges were also determined here.
The characterization of successful gripping processes enables the detection of faulty or unusual gripping processes, and a classification of gripping processes. “This enables the grippers not only to detect that a gripping operation has damaged a component,” explains Warnke, “but also to predict when the component is about to be damaged. In addition, the analyses have revealed information about the timing of different measured variables, such as the gripper current. This provides a valuable basis for making the grippers work even more reliably and a very good example of how AI can be used profitably – and without great effort in production and mechanical engineering.