Real-time Machine Condition Monitoring System for Predictive Maintenance

This technology offer presents an AI-based real-time classification and inference system to detect various machine operating conditions based on trained vibration patterns.

This solution uses Neuromorphic AI engine based on 3-layers artificial neural network with Radial Basis Function as activation function to model linear and non-linear vibration patterns. This solution operates without the need of a GPU, it incorporates a condition monitoring system and a low power edge-based AI.

This solution supports common types of condition-based monitoring including vibration analysis and monitoring, and temperature tracking. Real-time data is gathered through sensors, providing an ongoing method of testing and tracking machine health. Besides preventive maintenance, the solution also provides a predictive maintenance technique, in which the system spots upcoming equipment failure, and hence maintenance can be proactively scheduled when it is needed, and not before.

• AI based real-time classification and inference system to detect various machine operating conditions based on trained vibration patterns.
• Neuromorphic AI engine based on 3-layers artificial neural network with Radial Basis Function as activation function to model linear and non-linear vibration patterns.
• Does not requires a large training dataset. Ability to learn-on-the-fly (incremental learning) various vibration patterns while machine is in continuous operation.
• Detection of machine anomaly (report unknown vibration patterns). Able to learn (supervised) on-the-fly, any detected anomaly without the need to bring the system offline.
• Scalable system, ability to support multiple wireless accelerometers (or vibration sensors) across a large area using wireless mesh configuration for monitoring of multiple machines.
• Ability to adapt and use other sensors types such as sound and Acoustic Emission (AE) sensors.

1. Bearing conditions monitoring of a large motor

• Monitoring of rotational bearing health of a large electric motor pump

2. Tool conditions monitoring of a CNC machine

• A worn tool can increase the friction between the tool and the workpiece, affecting the machining quality and may cause time and production loss.
• Indirect method of tool condition monitoring via real-time acquisition of vibration signals during the machining process.
• AI-based prediction and inference of a tool condition.
• Monitoring from a CNC machine’s tool holder.

3. Anomaly (outlier) detection

• Anomalies can represent early signs of machine fault or failure, and are generally difficult to monitor or detect due to absence or insufficient information.
• By using Neuromorphic AI, unknown or out-of-range machine vibration patterns are treated as anomalies by default, which raise suspicions by differing significantly from the majority of the known vibration patterns.

• Extends the time between maintenance shutdowns. This is because maintenance is done on an as-needed basis.
• Does not require a large training dataset. The solution is able to learn-on-the-fly (incremental learning) various vibration patterns while machine is in continuous operation.
• Detect machine anomaly (report unknown vibration patterns). The solution is able to learn (supervised) on-the-fly, and detects anomaly without the need to bring the system offline.
• Scalable system. The solution is able to support multiple wireless accelerometers (or vibration sensors) across a large area using wireless mesh configuration for monitoring of multiple machines.