Industrial Process Fault Diagnosis in Case of Missing Sensor Data

Fault diagnosis is crucial for ensuring production safety and stability in process industry. However, missing sensor data is a common challenge that poses great difficulties in constructing and applying fault diagnosis models. Existing solutions rely on mathematical characteristics, such as the rank of the data matrix, to interpolate the missing values, and then input the model for training and diagnosis. However, these imputation methods fail to adjust to specific application purposes and often result in downstream tasks being treated as separate from the imputation process. To address this issue, this paper proposes a novel approach that considers the impact of fault diagnosis loss on data imputation to improve the accuracy of related variable repair, and ultimately, to develop a more precise online fault diagnosis model. The effectiveness of this model is validated using the multi-phase flow facility data set and real missing data from the oil well process data set.