Mathworks rolls out predictive maintenance tool

A TOOLKIT has been released to help engineers design and test condition monitoring and predictive maintenance algorithms.
Mathworks rolls out predictive maintenance tool Mathworks rolls out predictive maintenance tool Mathworks rolls out predictive maintenance tool Mathworks rolls out predictive maintenance tool Mathworks rolls out predictive maintenance tool

Toolkit helps engineers develop their own predictive maintenance algorithms.

Predictive Maintenance Toolbox from Mathworks, offers capabilities and reference examples for engineers designing algorithms to organise data, design condition indicators, monitor machine health and estimate remaining useful life to prevent equipment failures.

With Predictive Maintenance Toolbox engineers can analyse and label sensor data imported from files stored locally or on cloud storage.

They can also label simulated failure data generated from Simulink models to represent equipment failures.

Signal processing and dynamic modelling methods that build on techniques such as spectral analysis and time series analysis let engineers preprocess data and extract features that can be used to monitor the condition of the machine.

Using survival, similarity and trend-based models to predict RUL helps engineers estimate a machine's time to failure.

If there is only data corresponding to when a machine failed, survival analysis can be used to determine RUL.

Similarity-based models can be used when there is run-to-failure data available for the machines that captures how sensor measurements can change from healthy to failed states.

When there is data corresponding to condition indicator values over time along with information about critical threshold values that indicate failure for those condition indicators engineers can fit linear and exponential time-series models to their data to forecast when those thresholds will be crossed.

Condition indicators are features extracted from the data using time, frequency and time-frequency domain methods.

The value of those indicators typically change in a predictable manner as the health of the machine degrades over time with use.
The toolbox includes reference examples for motors, gearboxes, batteries and other machines that can be reused for developing predictive maintenance and condition monitoring algorithms.

The algorithms, which can predict equipment failure or detect underlying anomalies in sensor data, are developed by accessing historical data stored in local files, on cloud storage systems or on a Hadoop Distributed File System. Another data source is simulation data from physical models of the equipment that incorporate failure dynamics.

Engineers can extract and select the most suitable features from the data and use interactive apps to train machine learning models with those features to predict or detect equipment failures.

"Predictive maintenance is a key application of the industrial Internet of Things," Mathworks technical market manager Paul Pilotte said.

"This is critical to reduce unnecessary maintenance costs and eliminate unplanned downtime," he said.

"Engineers who typically don't have a background in machine learning or signal processing find design algorithms for predictive maintenance particularly challenging.

"Now these teams can quickly ramp up by using Predictive Maintenance Toolbox as a starting point for learning how to design these algorithms."