Building robust industrial machine learning (ML) models requires incorporating domain knowledge in feature selection. This ensures building meaningful ML models that fit the context of the industrial process that consists of complex networks of thousands of elements interconnected by flows of material, energy, and information. Despite the various automatic feature selection methods, they are still outperformed by the manual feature selection that embeds the industrial domain knowledge. This paper proposes an industrial feature selection method that (1) automatically captures domain knowledge from topology models holding information on the industrial plant and (2) identifies the relevant process signals (i.e., features) to a specified process element (i.e., to which an ML model is being built). We performed an empirical case study on an industrial use case to evaluate the effectiveness and efficiency of the proposed method in comparison to existing ones from literature.

An industrial process consists of a complex network of thousands of elements interconnected by the flow of material, energy, and information. Each of these elements can be attached to several sensors producing different process signals. Building robust industrial machine learning (ML) models requires handling this multi-dimensional process data while incorporating the process domain knowledge in the feature selection activity. Despite the variety of methods that automate this feature selection task, in industrial practice, they are outperformed with the manual feature selection by domain experts as embedding the industrial domain knowledge. In this paper, we introduce a feature selection method for industrial ML that (1) automatically captures domain knowledge from process topology models that hold information on the plant process and (2) uses it to identify the relevant features (i.e., process signals) to a specified process element to which an ML model is being built. We performed an empirical case study on two industrial use cases to evaluate the effectiveness and efficiency of the proposed method in comparison to existing ones from the literature. In the first use case, our method improved the ML model performance with accuracy of 95% and recall of 91%, compared to a baseline model that achieved 70% and 5% respectively. It also decreased the training time by 81% with a simpler model. In the second use case, our method competed equally with the other methods with regards to model performance, however, it outperformed them in decreasing the training time by 64% with a simpler model.

This paper describes an innovative approach and infrastructure for a seamless data communication between CAE and automation systems' engineering tools. In this approach syntax, semantics and graphical representations of various CAE tools can be captured, analyzed, visualized and mapped to an intermediate syntax and semantic in a data hub. Different captures from different data sources can be merged together and analyzed in order to enrich the data in the hub. Achieved results can consequently be shaped and mapped to any target syntax and semantics in a sandbox environment in order to be exported from the hub. This approach allows a bidirectional data flow from multiple CAE data sources and automation systems by utilizing a multi-level mapping library. All the intakes and outcomes of the hub are managed by a web based version control system which can provide the delta view between different versions regardless of their origin syntax or semantics. By one pass of data between the source and the target, the persistent bridge in the hub is formed and this bridge will be enriched and enhanced by each further pass of data. A prototypic development have been performed and the complete communication cycle have been verified using a sample P&ID and signal list from various CAE tools and an outcome of AutomationML format compliant to ABB's engineering tool as the automation target system.

Motor control offers a huge variety of area or custom specific solutions. These possibilities can be found in the process control system (PCS) also. However uncounted variants of device parameterization, different cyclic process image definitions and watchdog checks lead to a high complexity, which can be reduced by the consequent usage of profiles. Also today there are some profiles in use, but they are built from the field bus communication view point. This article introduces an approach which presents both, drive control and device diagnosis from the process control systems point of view, in order to reduce engineering cost. The suitability of the approach has been verified in a prototypical implementation.