Optimal sensor placement for identifying multi-component failures in a wind turbine gearbox using integrated condition monitoring scheme

• Published in: Applied acoustics 2022-02, Vol.187, p.108505, Article 108505
• Main Authors: Pichika, S V V S Narayana, Yadav, Ruchir, Geetha Rajasekharan, Sabareesh, Praveen, Hemanth Mithun, Inturi, Vamsi
• Format: Article
• Language: English
• Subjects: Condition monitoring; Grey wolf optimizer; Multi-component defects; Optimal sensor placement; Wind turbine gearbox
• ISSN: 0003-682X; 1872-910X
• DOI: 10.1016/j.apacoust.2021.108505

Wind turbine gearbox has a high failure frequency and downtime, and therefore, several sensors are installed to perform condition monitoring to reduce the operation and maintenance costs. A gearbox can have infinite sensor nodal positions, but, in reality, the positioning of sensors is limited to a finite number of locations. Also, sensor location influences the quality of the data captured by the sensors, which is of key importance in a condition monitoring system. Hence selection of optimal sensor placement (OSP) is a challenging task which needs to be addressed. When the sensor type changes, the measurement response changes, and hence the OSP methodologies based on the measured responses may not work well. For addressing this, an optimization method based on statistical features is proposed to find the optimum sensor placement (OSP). In order to evaluate the effectiveness of the proposed method, experiments are conducted on a laboratory scale model of wind turbine gearbox considering multi-component faults and an integrated condition monitoring scheme. Variational mode decomposition and the Spearmen correlation coefficient are used to processes raw acoustic and vibration signals. Feature extraction is performed to obtain nine statistical features, and a mathematical objective function is constructed as a function of these features. Grey wolf optimizer is employed to find the optimal values of the statistical features. Fault classification is performed using Random forest (RF) and deep multi-layer perceptron (MLP) algorithms. Optimal sensor network identified by the above method reported classification accuracy of 86.88% and 88.34% for RF and MLP, respectively. As a result, the number of sensors reduced from eight to five. The proposed method can be used as an effective technique for OSP problems.