This post is a summary of my talk on how Machine Learning is being used for the Predictive Maintenance of Heating, Ventilation and Air Conditioning (HVAC) assets. This talk was delivered as a part of TWIMLfest 2020.
What is Predictive Maintenance?
Heating, Ventilation, Air Conditioning (HVAC) equipment plays a very critical role in commercial buildings by maintaining thermal comfort and indoor air quality. The performance of a HVAC equipment becomes prominent when it malfunctions. So, maintenance is a major focus area for HVAC owners and operators. Maintenance is done either by replacing parts at regular intervals even when those are working (Preventive Maintenance) or by replacing the parts only when there is failure (Reactive Maintenance). Predictive Maintenance avoids the drawbacks of Preventive Maintenance (under utilization of a part’s life) and Reactive Maintenance (unscheduled downtime). Based on the health of an equipment in the past, future point of failure can be predicted in Predictive Maintenance. Thus, replacement of parts can be scheduled before the actual failure. Traditionally, predictive maintenance is being done using rule/domain based techniques. With the advent of connected sensors (IoT), data from HVAC equipment is continuously collected and fed to Machine Learning based systems to predict its future health. In this session, I am going to discuss how the HVAC industry is utilizing Machine Learning for the purpose of Predictive Maintenance.
Outline of the talk
- What is HVAC?
- Types of Maintenance in HVAC Industry
- Predictive Maintenance: Definition & Goals
- Architecture of Predictive Maintenance System
- Data Science Life Cycle for Machine Learning based Predictive Maintenance
- Conversion of Business Problem into Machine Learning Problem
- Collection of Relevant Data (HVAC Health, Maintenance, Asset Metadata)
- Aggregation of Data from different sources
- Data Preprocessing
- Feature Engineering specific to HVAC Health & Maintenance Data (Time Series)
- Selection of Cross Validation Technique
- Selection of the Algorithm based on stage of Machine Learning Pipeline
- Model Deployment: Batch Scoring
- Monitoring & Maintenance of the model
Arnab Biswas 