Manufacturing • June 29, 2023

Drive Industry 4.0 Transformation With Advanced Analytics For Predictive Maintenance

The maintenance department of a pharmaceutical company is given a tip - the industrial refrigeration system is going to fail tomorrow. The date of failure is divined through predictive maintenance, spurring the maintenance team to promptly schedule inspection and repair activities to avoid failure. On an industrial scale, predictive maintenance scenarios arise where historical and real-time data is collected over time from various parts of the operation to monitor the performance of equipment and processes. The objective is to find anomalous patterns that can help predict and ultimately prevent failures. 

Typically, when the cost of equipment failure is unaffordable, the equipment is subjected to preventive maintenance, which involves routine inspections and repair. Average or expected life statistics are considered to envisage when preventive maintenance is required. This differs from predictive maintenance where advanced analytics and machine learning techniques are leveraged to define the actual condition of the equipment and processes and to forecast their future states. 

Promise of Predictive Maintenance

Predictive maintenance provides an insight-led approach to optimize maintenance tasks in real-time, maximizing the useful life of equipment while avoiding disruption to operations. The implementation of predictive maintenance delivers a number of benefits and measurable results, as illustrated below:

The effectiveness of predictive maintenance as a maintenance strategy depends on three factors: 

  1. Collecting the right data: The quality of failure prediction improves with an increase in the amount of relevant data captured. 
  1. Framing the problem adequately: A comprehensive understanding of the kind of output required, recorded events, measured variables, and prediction horizon is needed to formulate the right modeling strategy for predictive maintenance. 
  1. Evaluating the predictions accurately: Based on the data collected, advanced analytics is leveraged to predict when failure is likely to occur and maintenance tasks are scheduled. 

However, before implementing a predictive maintenance model, it is important to assess if a particular operation is suitable for it. Applications that are suitable for predictive maintenance include those that:

  • Have a critical operating function
  • Have failure states that can be cost-effectively predicted with regular monitoring

Success Story - Predictive Maintenance in Oil & Gas Enterprise

The customer is one of the largest oil and gas companies in the United States,  with ten facilities in the Gulf of Mexico. According to the customer, approximately 7-8% of the facility downtime was attributed to malfunctioning compressors. The customer wanted to understand the cause of such operational failures and minimize faculty downtime using advanced analytics and predictive models. 

Quantiphi assessed the current landscape of the data architecture and noted a few preliminary challenges. The primary hurdle was to wrangle the incredibly large amounts of input data. Secondly, as per assessments, high throughput of streaming data was required with milliseconds of lag. In order to improve the performance of the predictive modeling, a third-party distributed, scalable Time Series Database application, or OpenTSDB, was required to be customized. 

As part of Quantiphi’s solution, the project comprised two primary workstreams:

  • A data pipeline from the customer’s on-premises software to the Google Cloud Platform that routes telemetry data to Google Cloud APIs, datastores, and visualization platforms
  • A data science workflow to facilitate predictive maintenance

Quantiphi built a Proof-of-Concept (PoC) solution with a keen focus on creating an end-to-end machine learning workflow for a single subsystem in a single facility using one machine learning model. After favorable outcomes were demonstrated, the PoC was scaled to work in tandem for three of the customer’s facilities against 17 subsystems using three distinct machine learning models. The solution was successfully leveraged to identify anomalies in their assets in advance, along with the cause of such patterns. 

Here’s a closer look at the business impacts of the solution:

  • $20 million per month in savings for 17 subsystems and 3 oil rigs
  • ~15 seconds near real-time inference of time series telemetry data
  • ~5 million records per second, autoscaled data ingestion
  • Prioritized safety of operators and reduced facility downtime

Get in touch with our experts to optimize your maintenance strategy with predictive maintenance. 

Written by

Aditya Sharma

Top Trending Blogs

Google Cloud

Prepare for Industry 4.0 with Smart Manufacturing

19 September 2023

Thank you for reaching out to us!

Our experts will be in touch with you shortly.

In the meantime, explore our insightful blogs and case studies.

Something went wrong!

Please try it again.