THANK YOU FOR SUBSCRIBING
I think we all know that operational excellence needs to be executed differently than 20 years ago because we need to transition to new energies, face a demographic change and need to be more efficient. Today industrial companies are hiring Subject Matter Experts for acoustics to find damages on industrial sites by ear, because they are lacking their own qualified personnel. Such manual anomaly detection is inefficient, costly and simply not timely anymore. Therefore, companies need to find new technologies that help running a reliable and efficient production and optimize maintenance cycles. AI and machine learning are some of those new technologies but how can they help? Can it replace an operator? And the most important question is: how to choose the right solution?
When I first did my research for predictive maintenance, I found that solutions were lacking built-in machinery knowledge and actionable insights on what to do next. The solutions weren’t flexible enough to add real value to operators. What they need is a deep understanding of what is happening inside the machinery and correlations within it.
Let me share how AI technologies are already making an impact for our customers and what the future holds for them.
For us the most important task was to develop a solution with a holistic approach that helps operators to understand what is happening inside their machinery. For that we decided to go for a hybrid approach that uses physical based models, also called first-principle models, combined with machine learning. This model exceeds just data-based statistical models, because the plant operators can understand the dynamical behavior of the machinery and understand correlations within it. Based on these principles the operator can simulate processes in the digital twin without intervening with the real process.
To complete a holistic monitoring approach of plants, it is necessary to not only collect the data from the sensors that are already available. By digitizing the information from inspection walks, operators have the data about the status of their plant and possible damages directly available in their dashboard. The digitization helps to understand current changes in the data instantly and improves the communication processes and workflows. For example, operators can directly create a maintenance order for the relevant personnel, decreasing downtimes of the machines.
Data availability is the key to a good AI solution. The better the data, the better the AI can learn from it and give better predictions about failures in the future.
The AI algorithm needs to become a self-learning tool that gets better over time. Therefore, the combination of the machinery data with the feedback from the operator is important. Only with the input and interpretation from operators who know the machinery well, it is possible to run the machinery at best efficiency. The retention of knowledge is most critical for plant operators. Information and expertise is lost when experienced experts are retiring and the training of new qualified personnel is time-intensive and without the expertise of retiring personnel impossible.
To give you an example, one of our clients uses our solution to monitor the compressor for their gas turbines in a large power plant in Berlin. The collected data was integrated into our analytics and the trained models were applied to the incoming data from the machine. Pretty early in the analytical work it became clear that the compressor was running frequently in an inefficient setup. That led to an increasing energy consumption. The energy consumption was increasing with minimal changes but resulted in an accumulated financial burden. With the use of our solution we could detect the anomaly and early give recommendations to correct the setup and retain the otherwise high energy consumption.
Operators are able to base their decisions on the newly acquired data, when operating the machinery. The acquired knowledge helps to prevent failures by having full access to the machinery data and seeing changes in real-time. Let me give you another example: a change in the vibration signal can be an indicator for possible damage. With the visualized data the possible damage can be detected, before the failure shows up in rising temperatures, which is the usual sign for damage. With a vibration analysis in real-time the prediction of damage is accelerated since vibrations are more sensitive than temperatures. If such an anomaly happens, our solution will notify the operator, which reduces the monitoring effort.
With the acquired data the AI algorithm is able, not only to predict damages and failures, but also to give recommendations on what to do next in terms of repair, maintenance and operational planning. A general practice in maintenance is to work in a periodic structure. In other words, the operator knows from experience that a machine breaks down every six months. He orders spare parts in a cycle of six months and changes the parts independently from their actual condition. A simple change of machine setups that would prolong the maintenance cycle or increase spare part quality and other factors are not calculated into the planning in a periodic maintenance structure and potential value is lost. With insights in the form of analyzed data the operator can make precise predictions about the machinery health state. With that knowledge, maintenance cycles can be planned flexibly and conducted when actually needed. The operator is then empowered to prolong or shorten maintenance cycles to either save spare part costs or reduce downtimes and guarantee runtimes. We have seen clients saving tens of thousands of Euros in repairs and spare parts in a year and extending on top the useful remaining life of machinery, for example from 20 to 25 years.
The decision for an AI solution is therefore not only a question of operational excellence and economic efficiency, but also a strategic investment into a future-proof company. It helps operators not only to have full access to the machinery data, but also to base the decisions on the data while operating the plant.