The use of data generated by the IIoT, combined with automation and machine learning, has provided a bright new future for physical asset management. By monitoring asset condition and degradation, these technologies are proving their value in eliminating losses caused by unplanned downtime, surges in capacity and other factors impeding efficiency.
Often due to regulatory requirements companies are looking to reduce their operating costs, increase their production or capacity, and output higher product or services to its customers at lower costs. Efficiency is the ultimate goal.
Data & Machine Learning
The volume of data being gathered every single day is colossal, and that volume is only set to rise even further as the IIoT gains momentum. However, data is nothing without the tools to decipher and draw insight.
Advanced, and largely autonomous, machine learning solutions learn behavioural patterns from streams of data gathered from, for example, sensors on and around machinery or assets, from mobile devices and computer networks. It is constantly learning and adapting to new signal patterns as operating conditions change, quickly ascertaining what constitutes normal and abnormal behaviours. Monitoring the condition and degradation of assets can highlight patterns throughout the lifecycle that can initiate alerts to take preventative measures in order to eliminate downtime or even potential failure.
The patterns and trends presented by the data can be handled at a much deeper level than is possible for the human mind. It can identify patterns that are invisible to the naked eye, ones which have never been discovered before, applying those patterns to new data.
For machine learning, the focus is on outcomes rather than the problems leading up to them. Those multiple, complex factors that do lead to the outcome are automatically uncovered, therefore creating a far more accurate predictive model. The more data that it is fed, the more is analysed, and the more accurate that predictive model will be.
What’s more, whatever the solution learns from one asset can, firstly, inoculate that asset to prevent it encountering the same issue again. Secondly, these learned signatures are then automatically transferred to similar assets within the network, inoculating them, too, against the issue.
These insights are automatically produced at a consistent rate, always with precise accuracy, saving valuable time and labour for the business.
Three types of Predictive Analysis
Essentially, the function of machine learning for assets boils down to predictive analysis, principally comprised of:
- Predictive Maintenance
- Demand Forecasting
- Workforce Management
These are functions previously assigned to humans. Automated predictive analysis improves upon human analysis, as machine learning is able to give twenty-four-hour cover, discovering insights almost immediately.
1. Predictive Maintenance
The first function, predictive maintenance, is one of the most applicable for industrial use of machine learning technology.
As part of asset information strategy, automated predictive maintenance uses data and models in order to predict when an asset will fail. This allows maintenance to be planned in a timely and appropriate manner.
A good example of machine learning use in predictive maintenance comes from Rolls Royce, who can monitor its engine condition throughout the flight predicting asset degradation ahead of lifecycle norms.
2. Demand Forecasting
It’s necessary for power companies to forecast for future consumption levels in order to balance the supply and demand in a cost-effective way. If machine learning techniques are used to analyse the data, it allows this to be done at a deeper level, taking into account correlations that were previously invisible. The result is the ability to ascertain hourly demand and peak hours, as well as more long-term predictions, thus optimising the power generation process.
Google DeepMind, for example, is working with the National Grid to predict demand patterns using its world-leading demand forecasting algorithms. Google believes it can knock 10% off the Energy consumption. In talks at the moment, but worth keeping an eye on.
3. Workforce Management
Workforce management, the third function of predictive analysis, is the business of optimising:
- Workforce productivity
- Labour costs
- Additional call-outs
By incorporating historical data on previous jobs, weather conditions, time of year, and other relevant factors, a machine learning system can ascertain where problems might arise. Equally, it is capable of scheduling staff to specific jobs based on their skill set, location, tools, and permits.
GE Digital acquisition of in ServiceMax last year shows exactly how critical it believes workforce management to be. GE estimates there is a market-wide opportunity to improve service productivity by $25 billion through the use of analytical tools.
All of this gives you some idea of the scale and scope of machine learning capabilities within asset-intensive industries. It’s clear to see the powerful benefits that machine learning has in predictive analytics across all connected assets, and how it’s set to change the face of the industry forever.
I’d be interested in hearing any other use cases for machine learning in asset-intensive industries? Please let me know if you have any, comments welcome.
If you’re interested in this topic, check out my article on how poor asset-data quality can undermine advances in analytics. I have also written about the uses of robots in asset maintenance.
I’d be interested in hearing any other use cases for machine learning in asset-intensive industries? Please let me know if you have any, comments welcome. You can contact me at [email protected]