Polymorph was brought onto an international consortium of companies to assist a German shipping pump manufacturer to meet 2 primary business requirements:
- firstly, generate recurring revenue for their business; and
- secondly, add continuous value for their clients.
It was determined that these business goals can be met by adding a predictive maintenance component to their business model as it could potentially transform their client’s business from only selling pumps to providing equipment-as-a-service.
As part of the consortium, Polymorph helped to develop and launch a scalable proof-of-concept model that eliminated the need to rebuild software once the solution is implemented on thousands of pumps.
Shipping pumps operate in complex environments inside large housings that are rarely visible from the outside. This makes it extremely hard to hear or see the typical warning signs or trends that predict an imminent failure or breakdown. It’s also important to keep in mind that effective predictive maintenance requires accurate data. Furthermore, the only reliable communication on ships is via satellite, which generally has a high cost and low bandwidth.
- Shipping pumps operate in harsh environments (high vibration, temperature changes, salt water, etc.), so sensors don’t last very long
- The only reliable communication on ships is via satellite, which has a high cost and low bandwidth
- Building a scalable Proof-of-Concept so that no software needs to be rebuilt when the solution is implemented on thousands of pumps
- Interpreting insights to take the right action at the right time
- There’s no historical data to train a machine learning model
A team of experts spent a substantial amount of time sourcing the right sensors that would be capable of withstanding harsh marine environments. Importantly, it had to process data offline and only sync the data when there was sufficient bandwidth. In addition, these sensors use a serverless architecture that enables horizontal scaling.
- Consists of sensors that have been proven to withstand tough marine environments
- Processes data offline on the ship and only syncs when there is enough bandwidth again
- Uses a serverless architecture that enables horizontal scaling
- User-friendly data visualisation that enables operators to take action
- Machine learning to predict machine breakdowns
- Off the shelf sensors
- Edge computing
- Wifi and mobile satellite data link to the cloud
- Amazon Web Services IoT Core
- Email notifications
Results / outcomes
- Launched a successful pilot in less than 6 months
- Spent a small amount of money and time to learn that predictive maintenance is viable and scalable (the pilot surfaced a lot of the risks early)
The predictive maintenance model was employed to proactively analyse and detect potential errors by, for example, the machine’s vibrations and so avoid pump breakdowns and the concomitant downtime. The ultimate goal is for this model to never stop learning. As it collects and processes increased data it becomes better trained to make better decisions.
The pilot programme was launched successfully in less than 6 months and emphasised the benefit of initially spending a small amount of money to first establish whether the planned predictive maintenance model is viable and scalable before rolling it out. During the pilot programme, risks and challenges would become clear early and could be addressed and corrected.