Application Of BP Neural Network On The Prediction Of In-situ Concrete Strength

Compressive strength is a key factor in the concrete quality control; it is the basisof structure design and construction, as well as the most important property ofconcrete. According to the Chinese Code, the quality of concrete structure members isevaluated by compressive strength obtained after 28d curing under standard curingcondition. Obviously, it is difficult to meet this requirement in modern construction,and it will result in an incipient engineering faults. Therefore, it is of greatsignificance to developing a technique using modern analysis method to predict theconcrete strength based on the early strength.Many theories and test methods have been put forward at home and abroad topredict the concrete strength at early age. Based on the principle of artificial neuralnetwork, the study was made on the input variables, network fabric, transfer functionand other parameters of the network by using MATLAB Neural Network Tool Box inthis work. Then a lot of concrete standard specimens were prepared in threesubstations in the middle-west region of the Inner Mongolia, these specimens curedunder standard curing condition were used as the training samples and testing samplesin this study. The basic BP algorithm, self-adapt algorithm with additional momentumand L-M algorithm were used to train the network. Based on a massive trial andemulation result comparison, a concrete strength prediction model was set up by usingthe L-M algorithm. In comparison with the multi-variant linear regression model, thismodel has a high accuracy and the errors can be limited less than 3%.