Induction Motor Fault Detection in ESP Systems Based on Vibration Measurements

Abstract

Condition monitoring of electrical submersible pumps (ESPs) is vital since they work in a harsh and confined environment. One of the most important parts of ESP systems is the electric motor which is conventionally the induction motor. Therefore, fast and accurate diagnosis of the motor in the ESP system can cut overhaul expenses and prevent further damages and costs. In ESPs, the vibration of the system is usually measured using the downhole sensors. Accordingly, this paper aims to propose a method for bearing fault detection in the induction motor based on vibration measurements. In this method, first, the best features of the vibration signals are selected using a wrapper approach based on the genetic algorithm. Then, the model is trained using the selected features based on k nearest neighbor classifier. The proposed method is evaluated using data of an experimental set from Case Western Reserve University (CWRU) Bearing Data Center.

Unsupervised constrained discord detection in IoT-based online crane monitoring

“…Vibration has been successfully used to detect anomalies in ball bearings of gear shafts [12], rolling bearing faults [21], bearing faults of submersible pumps [22], and a combination of ANN and SVM accentuated with acoustic and thermal imagery data of worm gearboxes [23]. Mukherjee et al demonstrate that a combination of vibration data from computer fans and a combination of CWT, PCA, and ensemble-based learning methods can identify the executing software processes in computers with a relatively high accuracy [24].…”

“…They tested the approach on a bearing dataset, demonstrating that the method outperforms other intelligent diagnostic approaches. Similarly, Alikhani et al propose a method for using vibration measurements from downhole sensors to detect bearing faults in induction motors of electrical submersible pumps by employing a genetic algorithm-based wrapper approach for feature selection and a k-nearest neighbor classifier for training [22]. Their method's efficacy is validated using experimental data from the Case Western Reserve University (CWRU) Bearing Data Center.…”

Locally optimized chirplet spectrogram for condition monitoring of induction machines in transient regime

“…The parameters α and β of these chirplet windows are obtained from (20), using (9) and (10). The resulting complex time window (15) is shown in Fig.…”

“…To ensure their reliable operation, it is necessary to detect their faults at an early stage [4,5,6,7], what allows taking corrective measures before the fault becomes a catastrophic one. Different condition based maintenance systems (CBMS) for IMs have been described in the technical literature [8], based on vibrations [9], sound analysis [10,11], thermal signals [12], etc. Among them, motor current signature analysis (MCSA) [13,14,15,16] has gained an increasing interest, because it can operate on-line, detecting different and possibly simultaneous types of fault, and can be implemented using non-invasive [17] low-cost current sensors, and fast processing algorithms, especially the fast Fourier transform (FFT).…”