Variational Mode Feature Construction-Based Improved Kernel Extreme Learning Machine for Rotating Machinery Intelligent Diagnosis

The complex operational environment brings challenges to vibration signal-based rotating mechanical equipment fault identification. On the one hand, the fault features under heavy background noise reveal weakness and nonstationary characteristics, which makes traditional spectral-based methods impuissant to extract effective features. On the other hand, the rotating equipment is served in normal condition for most of life, resulting in collected samples appearing small quantity and class imbalance. To deal with these issues, this article proposes a variational mode feature construction-based improved kernel extreme learning machine (VMF-IKELM) for bearing and gear fault identification. The methodology involves the following three steps. First, an adaptive variational mode decomposition (AVMD) is introduced to extract the nonstationary intrinsic features (IFs) from raw signals. Then, the typical indicators of IFs are calculated and reshaped to construct the variational mode samples to reflect the nonstationary characteristics from multiple aspects. Finally, the representative samples are input into VMF-IKELM optimized with particle swarm optimization (PSO) for rotating machinery intelligent diagnosis. Experimental study verified that this architecture can extract effective IFs and accomplish further high-precision intelligent fault diagnosis.

I. Introduction

As the core component of mechanical equipment, the quality of rotating mechanical components, such as bearings and gears, directly affects the safety and stability of production operations [1], [2], [3], [4], [5]. Therefore, research on failure mechanisms and fault diagnosis is essential to maintain the mechanical equipment operational efficiency [6], [7], [8], [9], [10].