A Data Quality Metric for Multi-Dimensional Data Sets of Sensor and Actuator Data in Process Automation

Data processing and Machine Learning have high potential to increase productivity and flexibility in automated production systems. However, sensor and actuator data often show deficiencies in data quality, in particular in the completeness of possible sensor and actuator value combinations. Following proven routines and processes during normal operations of an automated production system, leads to sensor and actuator data that is often limited to specific areas of the feature space. To assess the so-called Data Set Completeness a novel metric is developed measuring the utilization of the feature space in multi-dimensional data sets of sensor and actuator data. For evaluation, a data-driven condition monitoring approach for control valves in process industry is developed detecting the particular valve faults plug worn out and adhesions.
The proposed metric for Data Set Completeness is applied to the data-driven condition monitoring use case and the influence of Data Set Completeness on model accuracy of data-driven results is successfully verified.