The Research On Multisensor's Correlation And Intelligent Fault Tolerance For Truck Scale

Truck scale is widely applied to many fields such as storage, trade, transport, communication, industry and mine, which is the important branch of weighing apparatus. In existing truck scale, the outputs of all load cells in parallel are accumulated in connection box and then the voltage is obtained, which is in proportion to load's weight. By conditioning and A/D transition, the weighing result is gotten by MCU, and then it is displayed on indicator and transmitted to PC. There are two main disadvantages in existing truck scale. Firstly it is not easy to compensate its eccentric error and the accuracy of weighing result is low. In existing truck scale, the eccentric error compensation is realized by repeatedly regulating the potentiometer in junction box to adjust gain of each channel with load cell, which is fussy and labor-intensive. Secondly the existing truck scale is not fault-tolerant, thus it will be disable when any load cell is faulty.Under the supports of Ministry of Commerce of PRC Optimizing Composition of Importing and Exporting Electromechanical Device and High Technological Produce Foundation (Project [2007]301,"Intelligent Truck Scale based on Multisensor Information Fusion"), some studies are performed to improve the existing truck scale.The research status and development trends of truck scale and load cell are discussed, the composition of truck scale and their principle are introduced. The weighing principle of truck scale is expounded. The principle of load cells in series, in parallel and in series-to-parallel mode is analyzed and their characteristics are compared.The correlation of multiple load cells in truck scale is studied. By the radial basis function neural network(RBFNN), the relevant model of multiple cells'outputs and the relevant model of their outputs'ratios are founded. Then the global relevant model is established based on adaptive weighted fusion, and the output signal of load cell is accurately estimated. The simulated experiments show that the error of global relevant model based on adaptive weighted fusion is smaller than that of anyone relevant model and the arithmetical mean fusion.For truck scale's fault-tolerance and load cell's fault diagnosis, the failure cause and failure type of load cell is analyzed, and the model of intelligent fault detection for load cell is constructed based on the relevant model, load cells'fault features and detection criterion of voting fusion. Then many functions of model are realized when only one or two load cells are faulty in truck scale, such as addressing and isolating the faulty load cell, identifying the failure type, forecasting load cell's fault, estimating the faulty load cell's output and reconstructing the input vectors of weighing fusion network, etc. The simulated experiments show that these methods are useful.For expediently compensating eccentric error of truck scale and reducing the error of weighing results, the weighing model based on composite radial basis function neural network (CRBFNN) is constructed and the mechanism of its errors, such as eccentric error, linearity error, temperature error and creep error, is analyzed in detail. The CRBFNN model is composed of weighing fusion model based on multiple radial basis function neural network (MRBFNN) and truck scale's output fusion model. By using the CRBFNN model, the eccentric error, linearity error, temperature error and creep error are automatically compensated, and the accurate weighing result is obtained when there isn't failure load cell and there is only one or two failure load cells.The experiments show that the CRBFNN model is effective to improve truck scale's weighing.An intelligent truck scale is manufactured, which is composed by load receptor, load cells, the system of signal acquisition and process, the PC for training truck scale's model and management system of customer. The system of weighing signals acquisition and process is designed based on double CPU configuration (DSP and MCU), which is used to implement truck scale's intelligent fault-tolerance and weighing. The software for training truck scale's model based on LabVIEW and MATLAB is developed, which is used to get optimum parameters of models and download them to DSP. The management system is designed based on C#.NET Framework and SQL database, which is used to manage the weighing results. This truck scale is easy to operate and its weighing results are very accurate.The verification method for truck scale is presented, and then the performances of intelligent truck scale such as eccentric error, linearity, repeatability, discrimination and zero error, are verified in field, while its fault-tolerant performance is also verified. The error source of certified results is analyzed, and then its uncertainty model is established.The intelligent truck scale is verified in field by Liuzhou Institute of Metrology and Measurement Technology. Certified results show that the performance of the intelligent truck scale is better than that of scales with medium accuracy defined by Chinese National Standards"JJG539-97 Verification Regulation of Digital Indicating Weighing Instrument"when it is not failure. The global error isn't more than 0.7% when there is only one faulty load cell, which is lower than the design requirement (global error≤1%). When there are two faulty load cells, the global error isn't more than 0.8%, which is also smaller than the design requirement (global error≤3%). The batch production of the intelligent truck scale has started in Liuzhou.