Anomaly Detection in Predictive Maintenance with Time Series Analysis

Anomaly Detection in Predictive Maintenance with Time Series Analysis

Anomaly Detection in Predictive Maintenance with Time Series Analysis
How can we predict something we have never seen, an event that is not in the historical data? This requires a shift in the analytics perspective! Understand how to standardization the time and perform time series analysis on sensory data.

Most of the data science use cases are relatively well established by now: a goal is defined, a target class is selected, a model is trained to recognize/predict the target, and the same model is applied to new never-seen-before productive data.

The newest challengelies in predicting the “unknown”, i.e. an anomaly. An anomaly is an event that is not part of the system’s past; an event that cannot be found in the system’s historical data. In the case of network data, an anomaly can be an intrusion, in medicine a sudden pathological status, in sales or credit card businesses a fraudulent payment, and, finally, in machinery a mechanical piece breakdown.

In the manufacturing industry, the goal is to keep a mechanical pieceworking as long as possible–mechanical pieces are expensive – and at the same time to predict its breaking point before it actually occurs–a machine breakoften triggers a chain reaction of expensive damages. Therefore, a high value is usually associated with the early discovery, warning, prediction, and/or prevention of anomalies.Specifically, the prediction of “unknown” disruptive events in the field of mechanical maintenance takes the name of “anomaly detection”.

Read Also:
Don't count on data scientist being the sexiest job for much longer

The problem here is: how can we predict something we have never seen, an event that is not in the historical data? This requires a shift in the analytics perspective! If data describing normal functioning is what we have, then normal functioning we will predict!

For this project, we worked on FFT pre-processed sensor data from 28 sensors monitoring a working rotor. The FFT transform produces a matrix of spectral amplitudes for a time segment anda frequency value.

In order to standardize the time and frequency references, frequency values were binned into 100Hz-wide bands and time values were binned into dates. The FFT spectral amplitudes were then averaged acrosseach date and each frequency bin.Cells in the final FFT matrix refer to one single date and one single frequency band for a single sensor.Considering all sensors, we get 313 FFT spectral amplitude columns in total.

After this standardization task, a time alignment was performed, inserting missing cell values where no date was available. The final FFT matrix has dates on one axis, frequency bins on the other axis, and average spectral amplitudes as cell values, with occasional missing values.

Read Also:
Using OpenStack To Build A Hybrid Cloud With AWS

That is for each sensor and for each frequency band, we get a time series of spectral amplitude values evolving over time.

As our data set contains only data that describe the normal functioning of the rotor, we use these data to predict anomaly-free measure values and we measure whether such a prediction is good enough. If it is not, we can assume we are out of the range of “normal functioning” and we can trigger an inspection alarm.

Themore accuratethe prediction model for the normal functioning signal, the more precise and more robust the consequent alarm is that is triggered. With this goal, an auto-regressive (AR) model is trained on an anomaly-free time window using 10 past history sampleson each one of the 313 spectral amplitude time series.

Read Full Story…

 

Leave a Reply

Your email address will not be published. Required fields are marked *