Fault Diagnosis and Predictive Maintenance Methods for Robotic Arms

Robot Arm Fault Diagnosis and Prediction Technology Using Angle Prediction Error in LSTM seq2seq Models

This technology diagnoses faults by inputting multi-axis current sequences of a robot arm into a seq2seq model—comprising an LSTM encoder, a latent vector layer, and an LSTM decoder—to predict normal angle sequences and comparing the mean squared error against actual output angles with a threshold.

Existing model-based fault diagnosis struggles to identify failure mechanisms, while conventional data-driven methods face limitations in accurate prediction and diagnosis for multivariate systems where implementing physical damage models is difficult.

This technology proposes a method that monitors the error between predicted and actual angles in real time using a seq2seq model trained solely on normal current and angle data. Applicable to predictive maintenance in smart factories and industrial robot management, it enables early anomaly detection without the need for fault data, significantly improving equipment uptime.

‍

‍

Key Features:

• Training a seq2seq model to output normal angle sequence data by inputting normal current sequence data
• Configuring an LSTM-based seq2seq model that includes an LSTM encoder module, a latent vector layer, and an LSTM decoder module
• Predicting normal output angles by inputting current sequence data into the trained seq2seq model
• Diagnosing faults by comparing the error between the predicted normal output angle and the actual output angle against a threshold

‍

This invention was developed through the development of fault prediction and diagnosis technology for the Gyeongsangbuk-do smart manufacturing platform and the Ministry of Science and ICT's support for smart sensor-based intelligent building safety information in earthquake-prone regions.