Soft Sensor Change Point Detection and Root Cause Analysis

Society of Instrument and Control Engineers (SICE) Annual Conference, Kumamoto, Japan (To Appear),

, , , ,

[PDF] [Slides]

Soft Sensor Change Point Detection and Root Cause Analysis by Liang Cao, R. Bhushan Gopaluni, Lim C. Siang, Yankai Cao, Jin Li
Fig. 3 The framework of proposed method

Click to enlarge image.


Soft sensor has been playing an indispensable role in the process monitoring of key process variables. How to know if deployed soft sensor models are still performing well is a challenging but crucial topic for the industry. If there exists change points in soft sensor predictions, it indicates abrupt and significant changes in the process conditions. The presence of change points may require us to rebuild the model to ensure that it does not drift. Root cause analysis plays an important role in process monitoring when a change point occurs. Fast and accurate change point attribution is essential for timely recovery of model performance. This work proposes a straightforward way to detect the change points and find the root causes of changes. Off-line change point detection is used to detect changes by formulating change point detection as a discrete optimization problem. Then, we work on understanding which feature or combination of features that are shifting soft sensor predictions. Shapley additive explanations (SHAP) is adopted to explain the predictions of soft sensor model. It connects optimal contribution distribution with local explanations using the classic Shapley values. Finally, the effectiveness of proposed algorithms is validated on a real industrial data.

Read or Download: PDF

Can't find a paper? Create a GitHub issue to request a preprint.

DAIS Lab Publications

Read More: