Spartan’s Knowledge Transfer Partnership for asset management, PROPHES

 

I recently started at Spartan Solutions as a Knowledge Transfer Partnership (KTP) Associate, tasked with automating the training of the Machine Learning (ML) algorithms that power PROPHES, Spartan’s Intelligent Asset Performance Management and Predictive Maintenance solution.  In this blog, I’ll explain what the KTP scheme is and how it helps to drive the development of innovative services like PROPHES. I’ll then give some background to the problem I’ll be tackling at Spartan.

Knowledge Transfer Partnership

The KTP scheme is run by the UK government, to help UK businesses to innovate and grow. A KTP project involves three parties:

1. The company (Spartan) that forwards a proposal for a specific innovation project.
2. The academic institution (University of Strathclyde) with expertise related to the project.
3. A suitably qualified graduate, the “KTP Associate” (myself), who will lead the project at the company, with support from an expert supervisor at the University

From my perspective as an Associate, the KTP offers a chance to gain deep technical expertise while also developing the leadership and management skills that come with tackling such a complex project.

The government grant includes a generous stipend towards personal development, with Associates encouraged to attend courses and conferences related to the project area. And the link with the University means I’ve been receiving one-to-one mentoring from a machine learning expert. It’s been a busy but exciting first few weeks, and I’m looking forward to the challenges to come.

Intelligent Asset Management & Predictive Maintenance

In `asset-intensive’ industries, where operations rely on expensive equipment, downtime due to equipment failure can be extremely costly. For example, the breakdown of an offshore Gas Compression System will stop production on the affected platform.  The production loss can quickly run into millions of pounds.

It is desirable to minimise unexpected plant and equipment failures by performing regular inspection and maintenance jobs. However, this work is difficult and requires highly skilled engineers working in challenging environments.

There is a balance to be struck between minimising equipment downtime on the one hand, and minimising the number of maintenance jobs performed on the other. Two traditional strategies prioritise one goal at the expense of the other:

Run-to-failure: Operate equipment until it breaks down. Perform maintenance in a reactive manner to fix observed failures. The number of maintenance jobs is very low, but the downtime cost is very high.

Preventive maintenance: Attempt to minimise downtime by performing periodic inspection and maintenance. Downtime can be very low, but there will be a large number of inspections of healthy equipment.

With the advent of the Industrial Internet of Things (IoT), the physical properties of critical systems are increasingly being monitored by a network of sensors, which can feed real-time time-series data to remote data stores. The IoT enables the third paradigm for maintenance:

Predictive maintenance: Collect time series data and fault records from equipment. Use this data to train a machine learning (ML) algorithm to estimate the probability of an upcoming fault, given current sensor readings. Maintenance jobs can then be scheduled when this probability rises above a given threshold.

Accelerating Algorithm Development

Predictive algorithms are currently trained by data science experts, with support from rotating equipment engineers, using a laborious process that takes several months of the expert’s time.

Since both the observed physical properties and the possible failure modes will differ for each unique piece of plant and equipment, a new ML algorithm may have to be trained for each new piece of equipment the client wishes to monitor.

Spartan’s Intelligent Asset Management and Predictive Maintenance solution, PROPHES, supports a range of Artificial Intelligence algorithms, from expert rules to supervised ML.

Spartan’s strategic goal is to drastically reduce the amount of data science expert input required to build and train a new algorithm for industrial equipment.

ML algorithm development will be accelerated by building a new digital workflow to automate various time-consuming processes in the data pipeline. By removing these bottlenecks and reducing the costs of algorithm development, the PROPHES team will support the adoption of Predictive Maintenance in industry, making operations more efficient, and freeing up machine learning and data science experts to push the field forward even further.

It’s an exciting time to be working on the PROPHES project, and I will keep you updated on our progress towards full data pipeline automation.

To find out more about Knowledge Transfer Partnership (KTP)

To find out more about PROPHES