AI-Powered Predictive Maintenance — A Use Case


The AI & Analytics Engine helps manufacturers take proactive measures based on advanced data analytics to predict and avoid machine failure and optimize maintenance schedules.


Preventative maintenance involves the periodic inspection and repair of machinery assets, prior to failure, often determined by the time-in-service of the machine. There are multiple maintenance strategies considered preventative including; condition-based, time-based, predictive, and risk-based strategies.

Where the reliability of assets is mission-critical, the implementation of preventative maintenance strategies proves more cost-effective and efficient than corrective maintenance. The implementation of optimized and predictive strategies, in particular, is a multiplier of these benefits.

Using machine learning (ML), preventative time, condition, and risk-based maintenance strategies can be optimized, and predictive strategies implemented and productionized; Reducing down-times, improving run-times, and ensuring nothing is prematurely replaced, so the whole system stays functioning reliably and profitably for longer. In particular, ML can assist in;

  • Predicting remaining useful lifetime
  • Predicting failure within a given time window
  • Flagging anomalous behavior
  • Predicting failure probability over time and,
  • Optimizing maintenance schedules

Partnering with a South East Asian industrial chemicals manufacturer, The AI & Analytics Engine was able to predict failure within a given time frame by identifying the condition parameters that led to a recurring machinery ‘jam’ and alert the engineers with enough lead time to rectify conditions and reduce the impact to the business.

Maintenance Strategies

Predictive maintenance is a critical process for any industrial business and enables continuous, automated monitoring and just-in-time maintenance. This prevents potential downtime and costs caused by a reduction in productivity. Continuous Predictive maintenance also reduces unplanned reactive or corrective maintenance and reduces the cost associated with traditional preventive maintenance. Traditional preventive maintenance processes are often highly manual and require worst-case assumptions about equipment lifespans — This results in inefficiencies. Machine learning can assist in optimizing and automating some of these processes, increasing their efficacy and efficiency.


Productivity continues to rise due to modern machines. These machines are often highly complex and represent significant investments. Investments that are justified by their continued productivity.

Knowing ahead of time when an asset will fail avoids unplanned downtimes and broken assets. On average, predictive maintenance increases productivity by 25%, reduces breakdowns by 70% and lowers maintenance costs by 25%

This was the case for an industrial chemicals manufacturer based in Indonesia who needed to solve an operational issue they were facing: An increase in their factory equipment downtimes that they had no warning of prior to failure.

The figures from the factory revealed that downtime was steadily increasing week-on-week. Previously machine downtime had followed a consistent pattern and was largely manageable. Recently it was found that 2 of the 4 production lines experienced an increase of downtime by 6%, leading to a loss in productivity and revenue.

Manual inspection by the factory revealed this was due to raw materials coalescing within the mixing unit and jamming the delivery pipeline. However, it was not possible for technicians to rectify this “before the fact” because the relevant production parameters had been tuned previously, and were all within the “approved” working range.

The factory provided PI.EXCHANGE with 18 months of data readings from sensors placed at various points along the production line to solve this issue.


The company’s line engineers presented a detailed overview of the entire process flow across the stages the raw material goes through, from the beginning of the process till the end product.

The raw data was provisioned in a CSV format. This was ingested by the AI & Analytics Engine. Using the data wrangling feature, data preparation required minimal human intervention. Issues were solved with one-click solutions provided by the ai-powered recommendations. A recipe of actions performed on the data-set was created, so a repetition of actions could be performed on incoming raw data.

Once the data was cleaned, The Engine analyzed historical parameters from the production line data, and identified features that represented a combination of conditions that increase the probability of failure: The precise temperature of 3 (out of over 15) raw materials during production, combined with the time duration of mixing, was causing ingredients to coalesce and jam machinery.

From these features, The Engine recommended a prediction model to alert technical personnel to a possible downtime event, given the conditions.

The model prototype took 2 hours to develop. This process resulted in an offline prototype illustrating functionality and business value.

Next, the models were deployed on the production systems and integrated into the factory’s software environment via an API call. This enabled an integrated alert system for the technical personnel and a view of model life-cycle health, providing a sustained view of model fit for incoming data.

Try it for yourself with a 2-week free trial of the Engine!

Applicability Across Industry

Today poor maintenance strategies can reduce the overall productive capacity of a factory or plant. Long and continuous run-times of capital-intensive assets provide a competitive advantage, so too can efficient and strategic maintenance.

  • Applying the AI & Analytics Engine to assist in maintenance can help across:
  • Finding the most cost-effective times to maintain an asset
  • Identify possible cascade damage where the failure of one part causes wider damage
  • Providing a logistical edge, so all repair parts can be ordered and delivered ready ahead of time
  • Optimizing the scheduling of maintenance

The profit of predictive maintenance increases relative to the maintenance costs. The higher the cost caused by failure the bigger the benefits. If repairs can be carried out prior to failure, it is a opportunity to mitigate unplanned downtime, studies show unplanned downtime is costing industrial manufacturers an estimated $50 billion each year.

Where the reliability of assets is mission-critical, the implementation of AI-driven maintenance measures is increasingly important, industries include:

  • Manufacturing
  • Construction
  • Mining
  • Oil & Gas
  • Telecommunications
  • Power and Energy
  • Transport
  • Aviation

To learn about the use of Price Optimization with AI, check out this Use Case!

Interested in understanding how The Engine can help you and your industry? Book a demo with our team and see how you can implement predictive maintenance strategies!

Originally published at