Predictive Maintenance data analysis on SmartFactoryKL generated data

project-smartfactory-1

Based on the mutual correspondence between SDIL and SmartFactoryKL, we identified potential research fields for a cooperation that meets with both strategic research goals. The discussions led to a vision that is strongly related to the currently trending research topics “Industry 4.0” and “Internet of Things (IoT)”. The project intended by SDIL and SmartFactoryKL falls into the domain of real-time data integration and knowledge management. SmartFactoryKL will be used as a realistic research and demonstration platform. The SmartFactoryKL itself is a manufacturer-independent demonstration and research platform and unique in Europe. Here, innovative information and communication technologies and their applications are tested and developed in a realistic industrial production environment.

Manufacturing plants are characterized by a series of machinery that typically produce some goods during a series of sequential processing steps within a production line. Each of these steps makes use of different machines and tools that are customized for their specific task. The proper functioning of that used machinery, in turn, depends on its environmental parameters like voltage, pressure, temperature, etc. that are typically monitored on gauges, control lamps, etc. Modern machinery that satisfies the paradigms of the Internet of Things is characterized by a large amount of sensors that continuously offer such status information that is relevant to the production process. The sensors can be monitored, and alarms can be raised in the case of exceeding certain thresholds. Thus, the running production can be stopped until an engineer does some fixing. Moreover, since the nature of most sensors is much more complicated, their values vary over time even in the normal operation. A malfunction cannot be simply detected with exceeding thresholds. It rather requires time series analyses and recognition of outliers. There is, however, a larger economic potential for observing and analyzing these sensors. Even the slightest deviation of these dynamic values from the expected values can lead to a significant quality loss in the produced goods by so-called butterfly effects. An intelligent monitoring, storage, and analysis of the sensory data can have multiple positive consequences:

  • Analyzing time series of all sensors in a production line under normal conditions allows the training of anomaly detection algorithms. Such trained algorithms can classify the overall system performance much better than handcrafted rules for a selection of sensors and their expected set values.
  • Since only states classified as “not normal” need to be stored, the archiving of machine data for the purpose of later referencing and analyses can be much more efficient.
  • Since there is the possibility to not only use timestamps for indexing the data but also explainable conditions that led to the outliers, the retrieval of such extraordinary conditions within data logs for later analyses is easier.
  • Goods that were affected by outlier-conditions during their production can be identified by their serial number if serial numbers are used for indexing the (outlier) sensory data.
  • Live monitoring can detect outliers and thereby enable a much more efficient and faster intervention resulting in cost optimization and fewer substandard goods.
  • Long-term optimization of outlier detection: On the one hand, goods produced under outlier-conditions but without a significant defect in their mid-term life cycle can serve to accept their conditions as “normal” with the result of conceding a more relaxed anomaly detection algorithm.
  • On the other hand, goods produced under outlier-conditions with a significant defect in their early life cycle can help to define the conditions for a future production stop more precisely with the result of further cost optimization and reduced warranty cases.

project-smartfactory-2

The research project intended by SDIL and SmartFactoryKL is a prototypical study of the above-mentioned aspects. Since the sensory data is being provided by the production line of the SmartFactoryKL, it resembles real data containing the following information:

  • Module information, e.g., energy consumption, operating hours, production status
  • Product information (about order, client, priority)
  • Topology information (neighbors left/right, port ids)
  • Process information (process time, energy)
  • Transport information (transport status, errors, etc.)

The SmartFactoryKL production line consists of five vendor-specific modules serving for the manufacturing of the following exemplary product: a customized business card holder. The assembling process starts at the engraving module. Initially, all customized job data are written in the product memory. The product itself carries this information along the whole life cycle. The digital product memory is realized through a RFID chip that is integrated into the base plate of the product. The engraving module unloads the base plate of the business card holder and applies a customized engraving into it via a CNC-controlled miniature milling machine. Thereupon, the engraved base plate is rotated and placed on a workpiece that brings it to the next production module. The next two vendor-specific modules take over the mounting of a clip on the casing bottom of the business card holder and the assembling of the different colored casing elements. A robot places the customized cover on the base plate and force fits the parts together. Another module applies an individual 2D barcode on the topside of the business card holder through laser marking. The final module performs two different tasks: the final inspection of the product using a high-resolution camera, and the commissioning of the finished customized business card holder. Before and after each processing step, the production memory is read and updated via RFID. Various modules can easily be removed or added during the plant operation giving the plant operator the ability to select the manufacturing module of the provider that is the most suitable for the given requirements.

For the future development of such highly modular multi-vendor production systems, it is essential to monitor, control, and process data through all stages of development. A superordinate IT-systems needs to access data continuously for this purpose. Key technologies for enabling vertical integration within this plant structure are the communication standards OPC UA and an integrated web server. The integrated OPC UA server in every vendor-specific module guarantees the problem-free interaction of different modules beyond proprietary limits. Thus, a direct transfer of information (about the topology of the plant, workpiece-specific energy consumption, and status messages) to superior IT-systems is realized. All these real data containing the information mentioned above can be used by the SDIL. The broad application of future Industry 4.0 aspects will require new qualifications over the next few years. This project tackles predictive maintenance because it will be one of the key issues with regard to the future development.

With the IBM as a partner, the SmartFactoryKL plans to run a joint project with Smart Data Innovation Lab subdivided into the following three phases:

  1. By IT Summit 2015: The SmartFactoryKL gave ~20GB generated data to the SDIL that replicates a specific problem of the past. More precisely it is about a sporadic motor failure observed at several modules and finally identified as a result of the contact of the assembly belts of two adjacent modules. This scenario demonstrated by the SmartFactoryKL represents a quantitative extrapolation of the SDIL using the same analytics approach and technology at scale.

  2. By HMI 2016: In preparation for the HMI, the SmartFactoryKL will run endurance tests for the stability assurance of the facility. Massive data (and errors) will be generated and transferred to SDIL for further analyzing. At the end, the results will be fed back in order to improve the quality of the demonstration facility.

  3. The research will go on to identify different data gathering strategies, analysis techniques, as well as business models in the area of predictive maintenance, OEE, and quality assurance in the production context.

“Together with SDIL and the IBM Watson Foundation, we tested IoT potentials for smart factories of the future” says Rüdiger Dabelow, employee at the DFKI’s SmartFactory Laboratory. “The project lasted during spring 2015 and new possibilities of the internet of things for production and manufacturing were tried out.”

Data Innovation Community

Industry 4.0

Project Partners

Technologie-Initiative SmartFactory-KL e.V., DFKI, IBM

Contact Person

Plamen Kiradjiev, kiradjiev@de.ibm.com

Dominic Gorecky, dominic.gorecky@dfki.de

Project Duration

Nov. 2015 – Jun. 2016