Multi-frequency eddy current testing signal fusion using a complex-valued long short-term memory neural network

Eddy current testing (ECT) is widely applied to industry applications such as manufacturing and infrastructure condition assessment and maintenance. However, due to varied sources of interference associated with the inspection process, variations introduced to the acquired ECT signals may lead to the degradation of signal quality for defect characterisation. Conventional ECT signal processing approaches cannot address these issues properly, as they can hardly capture the complex features of signals efficiently and are less robust against noise and interference. Thus, this study proposes an ECT signal fusion approach using a complex-valued long short-term memory (CVLSTM) neural network. This method can enhance signal quality and improve defect characterisation capabilities by fusing signals at different frequencies. The experimental results demonstrate the advantages of multi-frequency ECT signal fusion in terms of signal quality metrics. Compared to real-valued networks, the complex-valued computations inherently preserve both amplitude and phase relationships, thereby producing fused signals with significantly enhanced quality. Thus, the paper highlights the potential of complex-valued neural networks for ECT signal processing and data analysis.