In order to deal with the problem that sensors in industrial production systems are prone to failure due to long-term exposure to harsh environments, a new method of sensor performance evaluation and fault diagnosis based on standard uncertainty is proposed. Taking the production process of square-tube as the research object, firstly, the magnetic scale and encoder are chose as the distance measuring sensor through the hardware test experiment, and the automatic adjustment system of the roll position is studied and designed, the roll positioning accuracy is improved within ±1 mm. Then, wavelet filtering is used to denoise the output sequence of the sensor, and the neural network is combined to predict the output sequence. Finally, the standard uncertainty is calculated by the residual of the predicted value and the actual measured value, the fault diagnosis of the ranging sensor and the evaluation of the real-time performance of the ranging sensor are realized by judging the size of the uncertainty. Simulation experiments on the faulty sensor with accuracy decrease showed that the fault detection accuracy of this method can reach more than 90%, and the proposed method is effective.