Return to Blog





Predictive maintenance solutions can eliminate unexpected downtime and losses in productivity, therefore saving millions of dollars in unexpected downtime. For that reason, many industries are actively evaluating the advantages predictive maintenance technologies can bring to a factory.

Today’s factories work in lines: Multiple machines operate in perfect synchrony to produce a variety of goods. Unexpected failures can put immediate stops to those lines, negatively impacting overall productivity.

While most factories perform preventive maintenance, failures still happen. Companies that desire to eliminate failures altogether should evaluate the prospects of adopting a predictive maintenance strategy, in which maintenance is performed when needed, rather than when advised. Employing such a strategy enables the use of predictive analytics to prevent failures and keeps unplanned downtime to a minimum.

The Proof of Concept

The first step in this journey is to critically analyze machine failures. That means going through historical maintenance records to identify reoccurring failures and determine which ones impact productivity the most.

Once those failures are identified, there are important questions to be answered, such as if those failures can be predicted, what risks exist with investing in predictive analytics solutions and what solution is most appropriate.

The best way to answer all of those questions is through a proof of concept (PoC). A proof of concept (or a pilot project) is a small-scale application of the solution a predictive analytics company offers. The company will evaluate items such as project feasibility, risks and advantages to determine whether or not the project will be effective. Companies can then decide which solution is worthwhile based on real-world validation.

Framework of a PoC
Framework of POC
1. Planning

In this phase, subject matter experts from both the customer and vendor meet to understand the overarching problem as well as the control systems deployed in the factory. This meeting enables them to identify the root cause of the problem and get acquainted with the available data, which could explain the failures. In this first meeting, the scope of the PoC is determined and a project plan is developed.

2. Implementation

This phase starts with parsing the data to the required format for further analysis. More often than not, clients provide more data than what is necessary to complete the project. In these scenarios, it is important to identify the most relevant data for the failure being predicted, after which data scientists employ several advanced machine learning algorithms to build predictive models that will accurately predict that failure.

3. Metrics Evaluation

Each predictive model built in the previous phase gives different results, but ultimately only one model is used when the solution is deployed. In this phase, specific metrics that indicate prediction accuracy are evaluated so the best model can be identified. In addition to accuracy, the time taken to generate the result is also evaluated, as a good balance between accuracy and evaluation time is necessary to predict failures in a timely manner.

4. Final Decision

In this phase, the clients evaluate the advantages of using predictive maintenance applications and analyze costs and return on investment for the solution. Overall user experience of the application is also evaluated and necessary customizations are discussed with the vendor so the final go-no-go decision can be made.

The PoC was a success. Now what?

The success of any industrial analytics project relies solely on how soon and accurately failures are predicted. In order for that to happen, the data collected on the factory floor has to be delivered to the analytics platform in a timely manner, which requires a data pipeline between the source and the application. Once a data acquisition strategy is put together, the solution can be deployed to the rest of the factory.

At Predictronics, our team of data scientists will closely analyze factory floor data to identify critical failures and develop a PoC that illustrates and proves the advantages of a predictive strategy. Our team will also demonstrate how PDX — Predictronics’ scalable, end-to-end predictive analytics solution for acquiring data, developing models and monitoring critical assets — can dramatically decrease unexpected downtime.

Interested in getting a proof of concept for your business? Contact us here.

Vinay Paladugu

About the Author

Vinay Paladugu is an Application Engineer for Predictronics Corp. with 3+ years of work experience in solutions engineering and industrial analytics. He assists Predictronics clients in the automotive, discrete manufacturing and transportation industries by helping them better understand and integrate new analytics tools into their production process and ensuring they have the intelligence they need to make data-driven decisions that positively impact their business operations. Vinay has a Master of Science in Business Analytics and Project Management from the University of Connecticut. Connect with him on LinkedIn.

For more information and weekly updates, follow Predictronics:

Know What Happens Next with PDX!