Raw sensor data fusion using Johansen cointegration for condition assessment of concrete poles

This paper presents a novel approach for raw sensor data fusion using Johansen cointegration, aimed at non-destructive condition assessment of concrete poles. The proposed Johansen cointegration-based signal fusion is compared with signal averaging, a conventional method, and the Adaptive Kalman Filter (AKF), an advanced signal fusion technique. These methods are applied to data collected from concrete poles under both laboratory and real-world field conditions, using an innovative narrow-band stress wave excitation system with a center frequency of 1 kHz. Our methodology begins with fusing raw sensor data, which is subsequently decomposed into narrow-band components, known as Intrinsic Mode Functions (IMFs), using the Variational Mode Decomposition (VMD) algorithm. From these IMFs, we extract a set of non-parametric and parametric statistical features based on Instantaneous Frequency (IF) and Instantaneous Amplitude (IA) signals. The results demonstrate the superiority of Johansen cointegration over both signal averaging and AKF in scenarios involving the high nonstationarity characteristic of real-world field data. Furthermore, the findings highlight a notable similarity between AKF and signal averaging, which may reflect the dominant linear properties in the recorded signals. We also propose an index based on normalized mutual information to facilitate a fair comparison with existing fusion methods. • A raw signal data fusion algorithm based on Johansen cointegration is proposed. • The method improves the robustness of the classification using statistical features. • The method is demonstrated on an NDT problem using in-situ and laboratory data. • For non-stationary signals, it is superior to alternative signal processing methods. • It is beneficial for in-situ data but ineffective for idealized laboratory data.