The IMU is widely used in the field of robotics due to its advantages of low cost, small size and sensitive response to dynamic changes. However, IMU has attitude drift during long-term measurement, which leads to inaccurate estimation of free acceleration, and thus affects its performance in anomaly detection. To address this problem, an industrial robot anomaly detection method based on kinematic model-guided IMU attitude error calibration is proposed. Firstly, the kinematic model-estimated attitude and IMU measured attitude are fused based on a Kalman filter to achieve the attitude error calibration for IMU drifting error in long-term measurements. Secondly, the free acceleration is calibrated based on the calibrated attitude and the calibrated free acceleration is used as anomaly detection samples. Then, the normal sample data distribution is fitted based on the Attentive Neural Process (ANP) to realize the reconstruction of the normal sample, and the anomaly discrimination is completed based on the reconstruction error. The experimental results show that the proposed method can effectively correct the drift error of IMU in long-term measurements. ANP can effectively detect the anomaly of the robot, and the accuracy and F1 score of the calibrated data improved by 0.58% and 1.26%, respectively, compared with the uncalibrated raw data.