Recovery of partial sensor failure for magnetic flux leakage sensors in pipeline inspection robots by block compressed sensing

Partial sensor failures are common for pipeline inspection robots in oil and gas pipeline inspection due to the harsh working environment. These failures usually give partial values that are beyond the normal range or keep unchanged for an unusually long time, which stops the latter signal processing. Current interpolation-based methods fail to deal with continues block failure. This paper proposes one plug-in sensor failure recovery method that does not require any hardware update. The core failure recovery idea is modelling the data of sensor failure position as data loss, and the data from health sensors contain information of it in sparse domain by modelling the problem with compressed sensing (CS). This paper designs a virtual Bernoulli measurement matrix based on failure data locations and treat health data values as compressed sensing samples for estimating the failure data. Block-by-block estimation scheme is used to make use of the high spatial correlation of a local region. The experimental results on real pipeline inspection systems show our proposed CS-based measurement correction method is robust to up to 25% of error rate, and it has better performance than the traditional interpolation method for most typical signal features, e.g. 36% improvement in 2D coefficients with baseline for spiral weld under 30% of error rate.