Design of fault diagnosis algorithm for electric fan based on LSSVM and Kd-Tree
- PDF / 3,324,722 Bytes
- 15 Pages / 595.224 x 790.955 pts Page_size
- 88 Downloads / 177 Views
Design of fault diagnosis algorithm for electric fan based on LSSVM and Kd-Tree Kongzhi Hu1 · Ming Jiang1
· Haifeng Zhang1 · Sheng Cao1 · Ziyi Guo1
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Currently, the complexity of mechanical equipment is increasing rapidly together with the poor working environment. If a fault occurs, how to find the fault in time becomes a poser. Motivated by this existing problem, based on the analysis of the fault characteristics of electric fans, a fault diagnosis algorithm model based on Least Square Support Vector Machine (LSSVM) and Kd-Tree was proposed. This algorithm was based on the LSSVM optimized by the Cuckoo Search (CS). This paper used the “one-to-many” principle and the sigma threshold method to introduce k-Nearest Neighbor (kNN) which was implemented by Kd-Tree as a secondary classifier to optimize the model. In data preprocessing, the data based on time series was first processed by Empirical Mode Decomposition (EMD) and the energy ratios were calculated, and the the above results were degraded by Principal Component Analysis (PCA) and normalized. On top of that, in case of the uncertain fault types, the Fuzzy C-Means clustering algorithm (FCM) optimized by Particle Swarm Optimization (PSO) was proposed to provide a priori knowledge for the model. In this paper, the algorithm model, FCM and other parts were verified to prove that the performance and generality of the algorithm were better than those of general classification algorithms, and relevant experiments were conducted for different data processing methods to expand the universality of the algorithm. Keywords Least Square Support Vector Machine · Kd-Tree · Data preprocessing · Fuzzy C-Means clustering algorithm · Model optimization
1 Introduction Nowadays, modern industrial production plays an indispensable role to the society due to its conveniences and efficiency. However, such complex mechanical production equipment may have unpredictable errors and faults during long-term and high-speed operation [1–3]. Equipment maintenance is both time-consuming and money-costly, and, worse still, fatal errors can even cause injury or death [4–6]. Therefore, it is necessary to implement a real-time diagnosis system to monitor whether an error is being generated or will occur on the mechanical equipment. Fault diagnosis has received widespread attention after the British doctoral scholar R.A.Collacott published “Structural integrity monitoring” [7]. Now it is further developed and applied, gradually integrating k-Nearest Neighbor [8, 9], Ming Jiang
[email protected] 1
School of Astronautics, Harbin Institute of Technology, Harbin, 150001, China
Support Vector Machine [10, 11] and many other machine learning classification algorithms [12, 13]. The introduction of BackPropagation Neuron Network (BP Neuron Network) introduces new ideas for fault diagnosis [14, 15]. More and more experts and scholars are investing in research. Such as Liang et al. proposed the use of Recursive Neural Networ
Data Loading...
