Making smart grids even smarter to revolutionise asset management in the electric power industry

Jon Lawson

Richard Irwin explains why combining machine learning with the IIoT and engineering models could revolutionise asset management in the electric power industry

The industrial world is awash with data and new information from sensors, applications, equipment and people. But the data is worthless if it is left untouched or not used to its full potential to gain insights and make improved decisions. To make the most of big data, utility leaders should implement machine learning alongside the Industrial Internet of Things (IIoT) and use 3D visualisation to take advantage of the increased insight they can bring to the operation regarding performance and reliability. Applied together and working in tandem, they can reap the rewards of cost savings and improved uptime. The smart grid is already transforming the industry, but with machine learning and reality models exploiting the IIoT, the smart grid has the potential to be even smarter.

Demystifying machine learning
We have all experienced some form of machine learning, from streaming movie recommendations to banks that monitor spending patterns to detect fraudulent activity. Now, the industrial arena is moving quickly toward using a type of artificial intelligence to leverage the IIoT.

As the velocity and variety of data becomes available through advancements in sensor technology to monitor just about anything, machine learning is being applied to efficiently manage increasingly large and fast-moving data sets. Previously, organisations with predictive analytics could use big data (current and historic) to try and predict future events with reasonable results. What it brings is a more accurate prediction using algorithmic models to deliver insight faster.

Machine learning can handle large and complex information, from sensors, mobile devices and computer networks, to discover hidden patterns or trends in the data. It can then learn these patterns and apply it to new, real-time data to detect similar patterns in the future. An example would be to model the performance of a piece of equipment, such as an overhead line, in relation to the temperature of its surroundings. Machine learning can be taught to see what normal and abnormal behaviour looks like, and by applying the model to current data, it can identify events, such as how, for example, internal and outside environmental temperatures will affect sag on the line. The system can then predict, from existing knowledge, that something isn’t right and send out notifications, and prescribe actions. The more data that is analysed, the more accurate the predictive model.

Part of the implementation process is understanding how it works and the number of techniques involved. Your software service provider or machine learning expert will recommend what techniques to use and when. The most common techniques are:
*  Supervised machine learning. The program is trained on a pre-defined set of ‘test’ data comprised of historical or similar data to the real thing, which then facilitates its ability to reach an accurate conclusion when given new data.

*  Unsupervised machine learning. The program is given a mix of data and must find patterns and relationships therein with no training whatsoever, without any specific target or outcome.

So, what it comes down to is knowing what it is that you want your data to tell you and understanding what sort of data you have available.

Unlike business intelligence and predictive analytics methods that require a significant amount of manual labour and time, machine learning automatically produces insights at a consistent and accurate rate. It can then apply the learning to new, real-time data for future predictions for easier and more reliable decision making.

 The continuous delivery of reliable and stable electrical power is paramount to utility companies. While users who operate on a 24/7 basis rely on a constant and an uninterrupted supply, it is imperative utility companies take every precaution necessary to reduce outages and downtime. In the electric utility industry, the ability to recognise equipment failure and avoid unplanned downtime, repair costs and potential environmental damage is critical to success across all areas of the business, as it directly affects the customer. This is even more relevant in today’s turbulent times affected by ageing assets, energy demand and higher costs. But, with machine learning, there are numerous opportunities to improve the situation. Some of the main forms of predictive analysis machine learning can deliver to the electric utility industry are detailed below.

One of the most applicable areas where machine learning can be applied within the utility sector is predictive maintenance. Predictive maintenance is the failure inspection strategy that uses data and models to predict when an asset or piece of equipment will fail so that maintenance can be planned well ahead of time to minimise disruption. Predictive maintenance can cover a large area of topics, from failure prediction, failure diagnosis, to recommending mitigation or maintenance actions after failure. The best maintenance is advanced forms of proactive condition-based maintenance. With the combination of machine learning and maintenance applications leveraging IIoT data to deliver more accurate estimates of equipment failure, the range of positive outcomes and reductions in costs, downtime, and risk are worth the investment.

Extending the life of an asset can be a low-cost alternative to capital replacement. With many utility assets nearing end of life, asset health indices can provide a safe and reliable solution to extend asset life as well as satisfy regulatory demands for proof of compliance and justify rate cases/budgets. Machine learning can improve asset health indexing methodologies empowering utilities to make defensible asset investment decisions. Even with a limited budget, asset health indexing software, such as that offered by Bentley Systems, is being leveraged to automate the analysis and ensure sustainability of the process.

The issue of imaging
Video and image interpretation is another issue. Thermal imaging has become a core predictive maintenance tool in any ongoing inspection programme. It is widely used for substation surveys and safety checks before planned maintenance work. This helps avoid costly service interruptions and equipment losses. Machine learning can be applied here to spot the patterns of the images of what a healthy piece of equipment looks like by identifying hotspots across transformers and transmission lines, therefore speeding up the time process.

Demand forecasting should also be considered when discussing machine learning. Accurately forecasting high levels of demand, such as within a utility service, gives a company a competitive advantage. It provides them with the information they need to meet customer demand by anticipating future demand or consumption. In the energy sector, storing energy is not cost-effective, so power companies forecast for future power consumption to efficiently balance the supply with demand. This sector is faced with twin problems associated with outages in peak demand, while too much supply leads to wasted resources. With advanced demand forecasting techniques provided by machine learning, utilities can ascertain hourly demand and peak hours for a day, allowing them to optimise the power generation process. Using information such as historical demand data, regions, population, weather patterns, events and so on, organisations can predict demand on any day or period in the future. This is essential for ensuring the utility can produce the resource required reliably and on time.

With the increasing use of smart meters in the home, energy companies can now tap into this data provided by the IIoT to give a more accurate picture of supply and demand by gaining more insight from individual habits. Machine learning can provide tailored insight into the user to inform utilities that energy use during working hours is low, for example, so they can offer more efficient thermostats.

Energy usage and load shedding
In businesses such as retail, where refrigeration accounts for nearly a third of the electricity bill, there are many techniques already in place that can be enhanced or even automated with machine learning. Areas such as load shedding, energy efficiency and alarm management will benefit greatly. For example, algorithms can be used to analyse alarms and identify which alarms are false and which are critical.

From these examples, we see how machine learning can be a benefit to both the organisation and the customer, for predicting outage and system failure, and long-term replacement decisions, accurate load forecasting at the meter/sub-meter level, better balancing of supply and demand, and detecting early warnings quicker, to providing tailored energy use, recommendations and reports, respectively.

As already seen, machine learning capabilities will help users realise insights from the large amounts of data provided by sensors and the IIoT. Bringing it all together is visualisation through engineering models for structures such as substations and plants. Engineering models or digital twin, are the computerised 3D model version of the physical asset, which maps everything associated to the asset using sensors to represent near real-time status, such as condition, performance and location. Where 3D models do not exist, users can quickly and easily create 3D models with technology such as ContextCapture, Bentley’s 3D reality modelling software. Here, using high-resolution photos, drones aid in the creation of digital engineering models of any structure. The photos are transformed into detailed 3D models of all infrastructure data – in a less labour-intensive, cheaper and more efficient manner when compared to traditional methods.

IT/OT convergence has become an accepted practice, with operators gaining new insight from known information. But misalignment in corporate strategy still results in silo building across many areas, especially within engineering technologies (ET), where engineering models often remain stranded, inhibiting the ability to leverage this information to optimise operations. They should be included with existing IT/OT conversation, driven by the IIoT as well as machine learning. Designing and testing new products, systems, and even plants in a virtual environment makes a compelling case, particularly from a cost perspective. Virtual models can tie these domains together over the whole lifecycle of an asset using its embedded digital DNA. From an asset management perspective, it’s about predicting a problem before it occurs and enabling maintenance to be performed at optimum rates and costs. This will be accelerated with the application of machine learning to make the decision-making process smarter, faster, and, more importantly, in context.

Continually modelling a substation, transformer, or tower means that personnel can survey the asset throughout its lifecycle, from initial design to current condition, applying the difference in data to maintain up-to-date information on the equipment’s condition along the way. These models become the context within which utility companies can design, build, and operate their infrastructure projects. Reality modelling can link engineers in the field directly to the office, sharing information and data collaboratively. Through the IIoT data provided by the images in the construction of the 3D models, machine learning algorithms will provide even greater context, a predictive capability and deliver more informed business insight to the user, resulting in faster and more reliable decision making.

Although machine learning gives the impression that human involvement is minimal, this is not the case. It gives the user more intelligence, context and insight to make the decision-making process easier and improve productivity. For those adding machine learning to their asset management journey, the next logical step is to go model-centric by adding visualisation dashboards, cloud-based IIoT data, analytics and reality models to machine learning. With a machine learning strategy in place it will give users unprecedented insight into their operation and lead to benefits in efficiency, safety and optimisation, as well as the speed in which decisions can be made.

With the arrival of the IoT, data is growing and becoming more accessible. With the ability to acquire more data, more advanced technologies are required to scrutinise and filter out the important information and the value held within. But, it can only be exploited by identifying what works well and what does not. Machine learning features complex algorithms to sort through large amounts of data, identifying patterns and trends within it, to make predictions. The use of machine learning in electrical utilities doesn’t have to stop at just transmission and distribution, but can be applied across the whole operation, where algorithms are used to continually improve overall performance across the whole facility and the equipment within it, and directly to the individual customer. By combining these machine learning practices with the IIoT and visual operations, they will bring, as they mature, significant benefits. The IIoT, engineering models, and machine learning should no longer be considered just buzzwords. Instead, combined, they should be a priority for achieving operational excellence.

Richard Irwin is with Bentley Systems

Recent Issues