Which constellations of machine data are typical for malfunctions and failures of a plant? And how can this knowledge be used to predict future problems? These are questions that Big Data and AI experts normally deal with in the context of predictive maintenance.
As part of a practical project on industry 4.0 at FutureLAB at Pforzheim University of Applied Sciences, these considerations were also on the timetable of some students of the bachelor’s programme in business informatics, management and IT last summer semester. In cooperation with the ERP specialist Asseco Solutions, they examined the data records of an Asseco customer from the aluminium die-casting industry and identified typical data constellations for machine malfunctions and rejects. The results of the data science project were presented in mid-July and now form the basis for the development of an AI-based system for preventive anomaly detection.
[Predictive maintenance techniques are designed to help determine the condition of in-service equipment in order to estimate when maintenance should be performed. This approach promises cost savings over routine or time-based preventive maintenance, because tasks are performed only when warranted. Thus, it is regarded as condition-based maintenance carried out as suggested by estimations of the degradation state of an item.
The main promise of predictive maintenance is to allow convenient scheduling of corrective maintenance, and to prevent unexpected equipment failures. The key is “the right information in the right time”. By knowing which equipment needs maintenance, maintenance work can be better planned (spare parts, people, etc.) and what would have been “unplanned stops” are transformed to shorter and fewer “planned stops”, thus increasing plant availability. Other potential advantages include increased equipment lifetime, increased plant safety, fewer accidents with negative impact on environment, and optimized spare parts handling. – Source: Wikipedia]
600 parameters in a single process
A good AI analysis requires a comprehensive database. However, even the largest big data reservoir cannot be used immediately if its data is captured without context. The aluminium die-casting specialist, who made his machine data available for the university project, also came to this conclusion. Already today, up to 600 parameters such as temperature, mould filling time or thickness of the press residue are recorded for each individual aluminium casting (“shot”). However, the data has not yet been allocated to specific faults or faulty shots. Accordingly, the existing data pool could not be used directly for analysis or error avoidance.
Students identify error-typical parameter constellations
The Data Science project, which was initiated by Prof. Dr. Joachim Schuler and Prof. Dr. Thomas Schuster from the University of Applied Sciences Pforzheim, focused on evaluating the existing data and processing it in such a way that useful findings can be derived from it. A group of nine students analysed the operating data of two aluminium die casting machines and developed a methodology to synchronise documented errors such as air inclusions, incomplete filling of the mould or insufficient hardening of the aluminium with the associated shot parameters. In this way, characteristic parameter constellations could be identified for the almost 30 defect types and assigned accordingly.
The results of the analysis were presented at the final presentation at the university in mid-July. The findings gained in this way flow directly back into the practical processes of the aluminium die casting specialist: among other things, it was shown that more than half of the errors were due to an incorrect configuration of the machine – a source of error that can be reduced with relatively little effort.
Basis for AI analysis
At the same time, the students have created the basis for the realization of a planned AI system for anomaly detection with their analysis. The AI experts at Asseco Solutions are currently using the identified parameter constellations as training data for a neural AI network, which is to be enabled to discover further, previously unknown critical parameter constellations analogous to the known patterns. By monitoring in the future whether parameters develop unusually and approach a critical error pattern, it will be possible to predict and significantly reduce both downtimes and rejects.
“The number of AI projects in industry will increase rapidly in the coming years – and so will the need for qualified specialist personnel,” says Steve Roth, Head of Operations at Asseco Solutions and head of the data science project. “In order to avoid a further worsening of the existing shortage of skilled personnel, it is essential to promote young researchers at an early stage. We wanted to make a contribution to this through the joint project with the students from Pforzheim University of Applied Sciences. For most of them, this was their first contact with the private sector – and nevertheless they approached the task very professionally. One student coordinated the process like a project manager, while others focused on the evaluation or on correlating the data with the errors. We were very impressed with this cooperation. As soon as the further steps of the project have been completed and the anomaly detection is actually running productively in the company, we want to invite the students again to our customer to show them how the results of their work are actually used in practice”.
Since 2016, the FutureLAB of Pforzheim University of Applied Sciences and Asseco Solutions have been working closely together in the field of teaching. In the past semesters, the Asseco experts taught the students the basics for the introduction and functionality of ERP solutions in practice in several lecture series. With the Data Science project, the cooperation has now been extended to practical applications. The project gave the students the opportunity to apply the theory concretely and to gain initial practical experience on future topics such as industry 4.0 and IoT.