The purpose of data fusion is to combine and process the data of multi-sensors, thus to obtain much more exact and reliable results than that of the single sensor. An improved data fusion method for commensurate sensors is presented in this paper. It overcomes the shortcoming of the traditional consensus algorithm with two sensors, which has different confidence distance while measuring in different precision. The relation matrix is fuzzified to avoid the subjective error in determining the threshold value. The results of numerical simulation shows that the improved method can make full use of the effective information from the sensors and it can help to improve the accuracy of measurement. Even when parts of sensors are affected or fully disabled, it still can get correct diagnosing results in damage identification.