Among the initial applications for digital twins in industry, predictive maintenance proved to be an early winner for the technology showcasing its practical use to avoid unplanned downtime. Industry analyst group ARC Advisory Group recently highlighted Owens Corning’s use of digital twins to improve operations and maintenance.
In addition to predictive maintenance applications, the utility of digital twins is also being proved in applications such as asset interoperability, new system simulation to reduce development costs and new product development. Another developing area for digital twin application is industrial energy savings.
According to Max Ivannikov, head of IoT at DataArt (an engineering services company that develops custom analytics and cloud systems), “By using data from sensors and other sources, the digital twin can simulate the behavior of machinery, a production line or a whole factory if required. This enables manufacturers to identify opportunities for improvements, including energy savings and optimize their processes to reduce consumption.”
Incorporating the use of artificial intelligence (AI) in digital twins to analyze large amounts of data collected for digital twin use can help identify energy use patterns. However, Ivannikov cautions that that the use of low-quality data with digital twins can result in inaccurate simulations and predictions.
To avoid this issue, it’s important to ensure the use of high-quality data that really affects the performance of the system instead of creating irrelevant digital noise. So, how do you do this?
Ivannikov says effective energy consumption monitoring combines several algorithms to increase the value of resulting data-driven insights. The most popular algorithms include:
- Thresholds
- Long Short-Term Memory
- Vibration Spectrum analysis
- Thermal Models
Algorithms and their practical application
The most straightforward approach to establish thresholds for various components is to focus on equipment aspects such as bearing temperature and motor vibration levels. “A sudden increase in temperature or vibration beyond normal levels can indicate potential malfunction,” Ivannikov says. “However, this method is effective only for identifying existing issues, and stopping the equipment immediately is necessary to prevent more significant problems.” AI algorithms help avoid the need for shutdowns as they can detect malfunctions before they occur.
“Long short-term memory (LSTM) is a popular method used to forecast the behavior of various processes,” says Ivannikov. “It is commonly used to predict a motor overheating, for example, without requiring a thermal model. Instead, we provide data to an LSTM model, which then identifies patterns of behavior that are expected to occur. By comparing current data to the model's prediction, we can detect any discrepancies and alert the maintenance team to check the affected machinery part.”
Ivannikov notes that, although the LSTM method is popular, “applying it to an actual motor is not straightforward. The equipment parts do not operate continuously and motors start and stop frequently based on the equipment’s mode and load. These parameters can affect the LSTM model's output, resulting in false alarms and missed detections. As a result, this method yields only average results but it is still a valuable tool for predictive maintenance in manufacturing, especially in identifying patterns of behavior for equipment that operates continuously.”
Monitoring vibration data is a long-established method for assessing equipment status. Ivannikov notes that applying a Fourier transformation to vibration data to identify low or high-frequency components is gaining a lot of attention. “The main challenge lies in collecting data with a high enough sample rate, which is necessary for accurate spectrum analysis,” he says. “To address this challenge, spectrum analysis can be implemented on a gateway device, instead of sending raw data to the cloud, which would be impractical and costly.”
Initial results from DataArt’s proof-of-concept studies indicate that spectrum analysis on the gateway device is effective in identifying existing malfunctions of machinery components, though it did not yield any patterns that could predict potential malfunctions in the future.
Thermal models have proven to be the most effective when it comes to energy savings analysis because heating is directly proportional to the electrical power consumed, says Ivannikov. “This approach has proven effective in detecting issues at an early stage, allowing for maintenance to be performed during scheduled downtime periods. The sensitivity of temperature changes in mechanical systems has made this method a valuable tool in identifying potential issues and preventing further damage.”
Ultimately, Ivannikov says the answer lies in combining different approaches to increase the chances of more accurate predictions and more effective energy consumption optimization. “Creating detailed models of basic elements such as motors, gearboxes, actuators and bearings of the system enables more advanced digital twin performance, leading to operational expenditure reduction and effective energy consumption.”
