Robust geophone string sensors fault detection and isolation using pattern recognition techniques based on Raspberry Pi4

Abstract

Abstract In the early detection of process faults, safety and the reliability of technical plants are two key drivers for continuous improvement. Detecting faults in Geophone string sensors (SG-10) is necessary in oil exploration to prevent economic losses. Methods are being developed to allow for earlier detection of process faults than conventional limit and trend checking based on a single process variable, and the development of these methods is a key issue. This paper analyses and compares the efficiency of pattern recognition techniques using a low-computational power device for drift fault detection in sensors, such as the Raspberry Pi 4. Support Vector Machine (SVM), Artificial Neural Networks (ANNs) and K-Nearest Neighbor (KNN) classifiers are among the pattern recognition algorithms being studied. The data for this study were from (SG-10). The four features (resistance, noise, leakage and tilt) were extracted from normal and faulty sensors in training data and testing data. Trained models have been offline tested, using models for sensor performance to detect drift faults. Specificity, sensitivity, and accuracy were used to compare algorithm efficiency, with 98% accuracy in the KNN, 98.3% accuracy in the ANN and 62% accuracy in the SVM. The results show that from the given classifiers the (ANN) and (KNN) outperform to the (SVM).