Usage of digital twin in order to predict temperature within a thermic test rig

Abstract: Many people suffer from diabetes and as a result of the disease, circulatory issues in feet are common. To find such issues at an early stage, Vistafeet were developing a product that measures the temperature in feet. This product needed to be calibrated and for that purpose, this master thesis was evaluating a proposed calibration rig and the possibility of using a digital twin in order to predict the temperature of the rig. The concept of a digital twin includes a physical product, a virtual model of the same product, and information flow between them. By receiving information from the physical product, the digital model should be able to estimate or predict information about the physical product, information that is then used to improve the physical product. In order to be a true digital twin, it should be automatic and in real time. If the data flow is delayed, a better description is digital shadow or digital model, depending on the level of connection. Due to time limitations during this master thesis, the real time connection required for a proper digital twin was not achieved. The scope was then limited from a digital twin to a digital model. The evaluation of the rig was conducted through a case study of the rig including a number of tests, with the purpose of replicating and verifying the result from a previous study by Xiao and Fan [23]. The digital model was made by logging data from the physical product to later use within the simulated environment. First the digital model was compared and adjusted to the information from several thermal points of the physical model. The thermal points were spread out to give as much information as possible the de simulation, but only placed on sides of the rig that would have easy access if the rig were to be used for calibration. Once the digital model was adjusted, the final simulation was made, and temperature data was compared in verification points to see how well the digital model fit the physical model. The verification points were chosen on the calibrating side of the rig and spread out to see if the model managed to predict potentially tricky places. To finalize the investigation of the rig, errors within the model and the rig itself were evaluated. The result showed that it was possible to fulfill the conclusions from a previously made study. The digital model turned out to be accurate and managed to predict the temperature down to ± 0,1 degrees for most verification points. However, one verification point close to a heater element had much less accuracy than the rest. The result was still acceptable, but this indicates that it is not possible to assume that the model can predict entirely correct temperature within the whole rig only because some points are correct. Especially if trying to predict temperature in more difficult places such as close to a heater. The investigation of errors within the digital model showed that the digital model simulated well within the limits as the temperature range and the controller changed. The sensor close to the heater remained in the same error range as in the first simulation. The errors in hardware were evaluated and the variation between sensors was measured to about 0,1 degrees. However, there might also be a slight offset from the true temperature due to errors affecting all thermistors equally. Even though a 0,1 resolution between sensors is quite good, it is insufficient resolution for this test since the errors in the tests were about ± 0,1 degrees. Despite that, the error of the simulation was still in an acceptable range for a digital model setup. For further improvement, a proper real time digital twin could be implemented, but also higher resolution sensors are required.