Detecting Anomaly in Chemical Sensors via L1-Kernel-Based Principal Component Analysis

Abstract

We propose a kernel-PCA based method to detect anomaly in chemical sensors. We use temporal signals produced by chemical sensors to form vectors to perform the Principal Component Analysis (PCA). We estimate the kernel-covariance matrix of the sensor data and compute the eigenvector corresponding to the largest eigenvalue of the covariance matrix. The anomaly can be detected by comparing the difference between the actual sensor data and the reconstructed data from the dominant eigenvector. In this paper, we introduce a new multiplication-free kernel, which is related to the ℓ 1 -norm for the anomaly detection task. The ℓ 1 -kernel PCA is not only computationally efficient but also energy-efficient because it does not require any actual multiplications during the kernel covariance matrix computation. Our experimental results show that our kernel-PCA method achieves a higher area under curvature (AUC) score (0.7483) than the baseline regular PCA method (0.7366).

The rise of automated curiosity-driven discoveries in chemistry

“…While anomaly detection methods have been extensively explored in areas like image classification and fraud detection, their application in the chemical sciences has been relatively limited. Two common and established use cases include the identification of faulty instruments 44 in chemical plants and sensor monitorization 45–47 in process control. In these cases, the methods are used on time-series data, where the event of interest is often underrepresented or absent in the training data.…”

Fault Detection via Occupation Kernel Principal Component Analysis

“…Fault detection methods for dynamical systems rely on the identification of anomalous behavior using measured data. Applications of fault detection range from healthcare [1]; manufacturing [2], [3]; monitoring sensor behavior [4], [5]; monitoring chemical processes [6], [7]; identifying the onset of nonlinear behavior in dynamical systems [8]; and identifying traffic anomalies [9]. A multitude of approaches to fault detection have been studied over the past few decades, such as data-driven, set-based, observer-based, and time-series analysis methods [3], [10]- [12].…”

“…Fault detection using KPCA / PCA relies on computation of a metric (T 2 , SPE, etc.) that measures how well new data can be reconstructed using the principal components [1], [2], [4]- [8], [15].…”