I. Establishing a Diagnostic Model Data fusion is divided into pixel-level fusion, feature-level fusion, and decision-level fusion. This paper adopts feature-level data fusion, the structure of which is shown in Figure 1. It can effectively utilize information while reducing information loss. Assume there are M sensors monitoring the operation of the system. Each sensor can extract different feature signals, such as voltage, frequency, and power spectrum. The input to the fusion center is: R = [S1, S2, ..., SM] Si [Si1, Si2, ..., SiN] where: Si is the output feature vector of the i-th sensor, Sij is the j-th feature value of the i-th sensor, and Ni is the feature vector dimension of the i-th sensor. Typically, Ni is different for each sensor. There is also some important information in the system that is not measurable; this information can be input into the fusion center in the form of observation and recording. The fusion center has different sensitivities to different feature vectors of different sensors. Different weights are assigned to different features of different sensors: wij is the sensitivity of the fusion center to the j-th feature value of the i-th sensor, and I is the maximum feature dimension among the M sensors. wij can be determined by expert scoring, statistical experiments, etc. For sensors with a feature dimension less than l, padding with 0. Data fusion reasoning methods mainly include Bayesian reasoning, evidence reasoning, and fuzzy reasoning. This paper adopts fuzzy reasoning technology, and its structure is shown in Figure 2. First, different feature values ​​are fuzzified to obtain a membership degree. Fuzzification can be performed using different functions. This paper adopts the normal function: μ(x) = exp{-[(xa)/b]2} (l) The fuzzy inferencer is composed of fuzzy reasoning rules. The general model of the reasoning rules is: Y = R·X (2) Y is the fault cause vector, X is the fault feature vector, and R is the reasoning matrix or inferencer. The key to fuzzy reasoning lies in obtaining R. Its general expression is: if (sl1, and s12) then f1 if (s21 and S22) then f2 and f3 if (s11 and s12 and s31) then f1 and f2 This form is easy to express expert knowledge and has strong interpretability. Finally, after defuzzification, the diagnostic results are output. Defuzzification methods mainly include the centroid method, maximum membership method, simple averaging method, and horizontal centroid method. Only the centroid method used in this paper is listed below: II. Application Example Taking a heating system as an example. The main test items include platform heating check, gyroscope accelerometer heating check, and temperature control circuit check. Three temperature sensors were used to measure the temperature of three different locations. The hardware connection structure is shown in Figure 3. The output signals of each sensor were fuzzified. The parameters of the fuzzification function were determined based on sensor technical specifications and expert experience. The fuzzification function for sensor 1 is: Based on expert experience and fault analysis, some inference rules were obtained, as shown in Table 1. The defuzzification function used was the center-average defuzzification function. Some experimental data and diagnostic results are shown in Table 2. As shown in Table 2, the diagnostic results for data conforming to the existing rules are completely correct. For data outside the rule range, the diagnostic results are also satisfactory, only the diagnostic membership degree is somewhat reduced. III. Conclusion The use of multi-sensor data fusion technology has the following advantages: 1. Fully utilizes sensor information, improving diagnostic reliability; 2. Fuzzy reasoning technology can be used to simultaneously diagnose multiple faults and fully utilize expert knowledge; 3. Rules are easy to modify. The rules of expert knowledge are independent of each other. When an incorrect rule is found, only the incorrect rule needs to be deleted or modified, and other rules are unaffected. Points to note in application: 1. Acquisition of expert knowledge. Expert knowledge plays a crucial role in the fusion process; incorrect expert knowledge will lead to incorrect diagnostic results; 2. The selection and extraction of sensor characteristic signals are the basis for diagnosis; the form of sensor signals should be fully analyzed.