Algorithms for sensor validation and multisensor fusion

摘要: Existing techniques for sensor validation and fusion are often based on analytical models. Such models can be arbitrarily complex consequently Gaussian distributions assumed, generally with a detrimental effect overall system performance. A holistic approach has therefore been adopted in order to develop two novel complementary approaches empirical data. The first uses the Nadaraya-Watson kernel estimator provide competitive fusion. new algorithm is shown reliably detect compensate bias errors, spike hardover faults, drift faults erratic operation, affecting up three of five sensors array. inherent smoothing action provides effective noise cancellation fused result more accurate than single 'best sensor'. Genetic Algorithm used optimise fuser design. The second redundancy on-line status output μH∈[0,1], where μH=1 indicates valid μH=0 when failed. This fuzzy measure derived from change detection parameters spectral analysis signal. scheme wide range fault conditions. An appropriate context dependent operator then perform competitive, cooperative or fusion, providing useful qualitative indication derive result. The operation both schemes illustrated using data obtained an array thick film metal oxide pH electrodes. ideal electrode will sense only activity hydrogen ions, however selectivity device worse conventional glass electrode. use reduce measurement uncertainty by combining readings multiple having responses. conveniently fabricated screen printing different oxides onto substrate.