[Part 1] How AI is Changing the IoT-Based Predictive Maintenance: An Introduction
Contributors Anurag Bhatia, Saurabh
Before we start discussing the role of artificial intelligence (AI) in IoT-based predictive maintenance, let me briefly introduce where and why there is a need for it.
Where It All Started
Manufacturing is a unique industry, one where many companies don’t know their actual infrastructure maintenance cost. They deal with the risk of unforeseen downtimes despite doing preventive maintenance. Therefore, industries have been increasingly relying on data science models to understand their operational behavior and the effects of preventive maintenance on expected lost costs.
Preventive maintenance was a technique previously used to prevent upcoming failures that force organizations into trade-off situations. But due to continuous evolution, machine learning (ML) models allow organizations to break the trade-off situation barriers by avoiding the unplanned downtimes and minimizing planned ones.
Predictive Maintenance Problem Formulation Strategies
A predictive maintenance problem can be formulated in many ways, but the following seems to be building blocks in problem formulation and outlining.
Failure status could be observed using classification and anomaly detection algorithms that run on raw and feature engineered datasets from sensors.
Datasets used in these systems need to go through a rigorous exploratory data analysis process and were shaped over time by feedback from maintenance engineers.
Final alerts can be shown in the dashboard or sent over email/SMS to the facility maintenance personnel.
The Complexity of the Predictive Maintenance Problem
Complexity is a multi-faceted phenomenon, involving a variety of features including disorder, nonlinearity, and self-organization. The complexity measurement parameters are listed below in order (with the probable outcomes they may change) but are mostly measurable.
Entities, in our case, could be sites, site locations, or sensors.
Transactions can be from the sensor output, maintenance logs, and telemetry records.
Disorders could be location-based, operational, maintenance driven, or lack of maintenance, monitoring misconfiguration, due to parts failure.
The system under maintenance can be of various types such as mechanical, electrical, software, or hardware.
A client’s (whether internal or external) role is an important condition in various areas of problem complexity such as domain knowledge, alert relevance, and system usability.
To measure complexity, we need to apply the conditions defined above and analyze the iterations used to achieve the robustness, acceptability, and modularity of the application. This is achieved on optimized computing resources.
What’s Next?
Let’s take a deep dive into the datasets, models, inference, evaluation, and optimization in the upcoming stories:
[Part 2] How AI is Changing the IoT Based Predictive Maintenance: Datasets and ML Models
