Case study: Polymorph assists international manufacturers with predictive maintenance using advanced IOT and machine learning

Operational downtime is one of the biggest risks that large manufacturing companies are exposed to. The business cost of a critical machine failure and downtime can cost millions.

Introduction

For companies where cost of critical machine failure and operational downtime is a big risk, countermeasures to avoid downtime is crucial. In this regard, one of the best countermeasures to avoid costly downtime is predictive maintenance. Predictive maintenance offers a proactive approach to solving unplanned interruptions due to equipment or machine failure that can lead to costly downtime.

Polymorph is involved in the development of two predictive maintenance solutions founded on IOT data and incorporating machine learning. The first is an international German shipping pump manufacturer and the second is for the predictive maintenance of large industrial solar energy plants.

Richard Barry, CEO at Polymorph further mentions that the company is “investigating another opportunity in mining, to optimise their processes while implementing a predictive maintenance solution. Mines are very dependent on, for example, the commodity prices and operate on very low margins. Utilities and mining often have data, but it is not usable as it resides in different places. It is captured in different systems in various excel spreadsheets. Mining operations usually have ample data but are not utilising it optimally and often only receive the data when it is already too late. We start by analysing their existing data and see how much we can learn, gain insight, and then add and improve.”


The challenge

German shipping pump manufacturer

Pumps operate in complex environments inside large housings that are rarely visible outwardly which means it is often harder to hear or see the warning signs or trends that predict an imminent failure. In this scenario, predictive maintenance can proactively analyse and detect potential errors by, for example, the machine’s vibrations as a result of the ocean hitting against the ship as well as outside temperatures. The predictive maintenance model can further eliminate environmental noise and so render diagnostics more accurately. A further benefit is that the ship’s management team is now able to understand both current and future machine functioning abilities. With proper predictive maintenance, over the lifetime of a machine, faulty components can be swapped before malfunctioning or performance decreases. It also opens the possibility of an alternative business model for the pump manufacturer, allowing them to move from selling pumps to providing Equipment as a Service.

Energy plant company

A photovoltaic (PV) solar plant supplies electricity through a complex combination of equipment, including many solar panels. The output of the plant is influenced by factors such as the weather and the direction of the sun. Failure of the equipment can happen on a small part of the large, complicated plant and can go unnoticed for a lengthy period of time. This, again, is where the benefit of predictive maintenance is evident – it has the ability to detect faults or potential breakages before the human eye detects it and, of course, before a costly breakdown occurs.


How Polymorph addresses the challenge with machine learning


Polymorph was granted access to the manufacturers’ existing and historical data to first analyse and then use to create and improve a machine-learning model that can detect operational anomalies and determine on a granular level when and exactly where certain equipment is starting to drift outside of its performance specifications which is usually indicative of an imminent breakdown. In respect of the solar plant, it measures a whole host of data points, for example voltages and currents inside the entire grid.

Next, the machine learning model analyses all the data and determines if the machine or piece of equipment functions optimally. It learns from historical data and, over time, creates a trained model using and applying data such as the time of day, month, and year comparable to applied statistics.

Similar techniques are used in, for example, fraud detection on credit card transactions. The trained models can detect the fault quicker or identify a problem better than humans.

The Polymorph Platform


It is important to keep in mind that effective predictive maintenance requires accurate data. The data, however, is often found at various locations and not optimally utilised. A further important aspect to consider is cleaning the data and so exclude any unnecessary data that plays no role in improving and refining the machine learning model.

In terms of data storage, data is stored in a central cloud database making it easy and fast to extract and write to. Polymorph further created an easy to use user interface that can be built into intelligent dashboards. The next step in the process is to build a trained, healthy model that will be used to detect errors and abnormal behaviour. The aim of this model is for it to never stop learning. As it collects and processes increased data it becomes better trained to make better decisions.

Alerts and reporting

The Polymorph platform can easily integrate with business processes, create alarms and send emails and SMS’s to selected parties in case of an emergency or simply inform the maintenance team should a machine or piece of equipment need to be serviced.

Conclusion

Downtime as a result of machine failures can be more costly than the actual machines breaking and, as mentioned above, can cost companies millions. Predictive maintenance using advanced IOT and machine learning can prevent unwanted downtime and save organisations a lot of money. Machine learning is a cost-effective approach to predicting imminent failures and the maintenance cycles of machinery or equipment. This technique detects any deterioration or the onset of it in machinery that requires maintenance and it can aid energy plant- or ship pump- manufacturers to address any maintenance issues beforehand.

Please contact Polymorph if you have any questions regarding predictive maintenance. Polymorph focuses on ensuring the technology always supports the best solution.

“Downtime as a result of machine failures can be more costly than the actual machines breaking and, as mentioned above, can cost companies millions. Predictive maintenance using advanced IOT and machine learning can prevent unwanted downtime and save organisations a lot of money.”Richard Barry, CEO at Polymorph