We address automated production equipment in a wide variety of industries. Despite the diversity of technical applications with a high variety of variants, there are still many similarities in existing processes as well as in control systems and data. Our generic approaches and methods thus create added value for a wide range of technical applications.
Complex processes are usually controlled by numerous, linked instances. We enable for the connection to different data interfaces. Supported data interfaces include current machine data protocols, various fieldbus systems and wireless standards.
We realize connections between operating and information technology by means of state-of-the-art IIoT and Edge Gateway technology. Data is preprocessed locally and preconditioned for desired purposes. Relational and nonrelational databases as well as analysis platforms are implemented on-site, in corporate data centers or the public cloud. Our modular software services are designed to be portable and independent of hardware.
The generic algorithms for data driven training of process models require discrete event flags as well as continuous data series. Discrete data is taken from internal state machines as well as control and status words from existing control systems. Feedback-controlled actuators or sensors for process and inline quality control provide continuous measurement data at regular intervals. Where necessary, additional sensors can be seamlessly integrated.
Robust handling of distributed and heterogeneous data sources over stable data stream pipelines presents a significant challenge in general. To unify the characteristics of complex systems, including their running processes, requires precise data synchronization. The preprocessing steps may include further measures for data cleansing, anonymization, encryption and compression as well as extraction of relevant signal features near the data source.
The purely data-driven training of the process model initially includes an automatic segmentation of different process states as well as the learning of valid process step sequences. The resulting, interpretable timed automaton accurately maps the temporal behavior including occurring variances. Continuous time series data are mapped for individual process steps in machine learning models.
The process visualization initially creates the highest level of transparency about the application considered. Online monitoring provides deep insights due to the structured scope of the complex data foundation. In addition to the (remote) monitoring of running processes, the possibility of continuous recording of the process integrity and other KPIs ensures any required traceability or benchmarking.
In comparison with the learned model even the slightest anomalies of the ongoing process can be detected in real time. In addition to deviations in the durations of individual process steps, statistical anomalies in the course of existing branches as well as unknown events, states or state transitions are detected and trigger reports and alerts. Deviations of the continuous time series data indicate wear, overload, vibration or system detuning.
The detailed image of the plant and process behavior enables the analysis with regard to existing optimization potentials. Possible measures include the review of control programs (code re-engineering) to eliminate unwanted plant behavior, the statistical analysis of recorded process sequences and the detection of cycle time potentials or the reconstruction of energy flows to increase system efficiency.