Jump to content
  • Unplanned Maintenance


    This article demonstrates how to build a classification model, using Spotfire Data Science - Team Studio, for predicting equipment failures that would lead to unplanned maintenance. This use case can be extended to multiple different industries and scenarios, including heavy machinery, electrical equipment, and even some IT failures

    Use Case Overview

    This Playbook demonstrates how to build a classification model for predicting equipment failures that would lead to unplanned maintenance. This use case can be extended to multiple different industries and scenarios, including heavy machinery, electrical equipment, and even some IT failures. The common tie between these possible use cases is the consistency of test data intervals, outlined below.

    Data Requirements

    We use sample data from a mining company to illustrate the concepts in this Playbook, encompassing two years' worth of automatically-generated test data at 10-minute intervals, as well as a set of maintenance orders describing equipment failures. The features in this Playbook include metrics such as motor speed, tank pressure, and temperature readings from RTDs (resistance temperature detectors). Your dataset does not need to follow these features exactly, but it is important that any type of test metrics used follow regular time intervals, without missing values.

    Data Preparation

    data_preparation.png.e70f033b663ded23340aab230585fb91.png

    First template workflow for data cleansing and feature engineering.

    Our first step will be to clean the dataset and create features that support our unplanned maintenance model. We start by blending together three main sets of test data, all recorded at 10-minute intervals:

    1. Temperature readings from RTDs

    2. Runtime metrics from the motors (such as current and motor speed)

    3. Liquid flow and density measurements

    We blend these metrics together based on the time stamps associated to the recorded tests. Then Window Function operators are applied to aggregate data based on various time intervals, in this case ranging from 1 to 24 hours before any given timestamp. These aggregated measures will serve as features in the modeling process defined below. 

    To identify points of failure in the system, work order logs are joined to the feature set, accounting for any rounding necessary to map onto the original 10-minute test intervals. A variable operator is used to create a binary classification label, unplanned_maintenance, where work orders representing unplanned maintenance that require shutdown of production equipment are assigned a 1. If desired, failure data and features can be filtered to specific parts of the machinery - here, we use row filters to split data for part 130A and part 130B.

    Model Building

    model_building.png.6045fb9a2ea11b5194a80aa34f54a461.png

    Second template for predictive modeling.

    The final step is to build a classification model on equipment runtime features that predicts those times and machines where unplanned_maintenance is 1 (machine likely to fail in the next 24 hours). For this step we do some final transformations (replacing nulls and outliers), then apply four classification techniques to our training sample, which we've resampled to over-represent equipment failures.

    Key Technique - Correlation Filter

    Several of the variables in the machine maintenance dataset are highly correlated with one another, such as motor speed and motor current. The Correlation Filter operator automatically identifies groups of columns that are correlated with one another past a configurable threshold and filters all but one. The remaining column is the one most correlated with the dependent variable (unplanned_maintenance in this case).

    In our analysis on the mining equipment data, the Alpine Forest model was best at predicting upcoming failures, with 85% accuracy on the hold-out set. The accuracy and class recall could be improved with larger volumes of failure data. We provide a variety of other evaluation measures and export this high-performing model out to the workspace as PMML.

    Check It Out!

    For access to this Playbook, including its workflows, sample data, a PowerPoint summary, and expert support from Spotfire® Data Science data scientists, contact your Spotfire® Data Science sales representative.


    User Feedback

    Recommended Comments

    There are no comments to display.


×
×
  • Create New...