MJP Dicision Fusion Method And Its Aplication In Structural Damage Detection

Decision fusion is extensively used in multi-sensor data processing because of its merits, such as few communication data, strong fault-tolerant and low request for homogenesis of sensors. As the major decision way, decision fusion method directly affects the performance of entire decision system. In order to break through the limitation of the existing decision methods, and enable the fusion center to fuse all kinds of local decisions to obtain a more accurate global decision under especial or common cases, maximum joint probability (MJP) methodology and its adjustment for different decision circumstances were proposed in the view of probabilistic decision based on Bayesian decision theory. Combined with numerical simulation and model experiments, a rather complete theoretical study on MJP methodology was conducted.After generalizing the common properties of decision fusion with multiple hypotheses, a general mathematical model was built. Based on the model, MJP methodology was theoretically proved that it has the ability to get minimal error probability when the probability of hypotheses and local decisions are known. The effect of the factors, such as number of hypotheses, number and correct probability of local decisions, on correct probability of global decisions was summarized.Based on the concept of MJP network, adaptive MJP decision fusion method was presented, which enables the fusion center to fuse local decisions similarly when the probability of local decisions are unknown or changing, through unsupervised learning, calculating weight automatically and adapting the change of circumstance gradually. After proving the convergence of adaptive MJP methodology by the way of theory and numerical simulation, adjustment method for updating rate of weight was offered.To make decision fusion be more suitable for structural damage detection, multi-precision and multi-damage MJP methodologies were proposed, which enable the damage detection system to fuse detection results with different precision classes, such as damage existence, damage location, damage level, to get more accurate final decision when multiple damage appears at the same time probably. In order to make the damage detection result offer more information for structure maintenance,...