Implementing Virtual Sensor Technology for Aero-jet Engine Test Environments

Background and Objectives

Aero-engines are complex mechanical systems that supply the propulsion component of aircraft. Aero-engine development and optimization necessitate a large investment in research and development as well as testing. Sensor-based monitoring of engine components, as well as data analytics, can be used to track component degradation and automate mechanical subsystem diagnostics. The use of these automation technologies in Engine Test Environments can drastically cut engine development costs while also improving the performance and dependability of developed systems. Virtual sensors are utilized to solve the problems that physical sensors have, with cheaper running costs and increased dependability, as well as the capacity to measure non-measurable qualities indirectly. IVA has been working on the GTRE GTX-35VS Kaveri for Aero-Engine Test Environments that handle component subsystem dependability. The Defence Research and Development Organisation (DRDO) in Bengaluru, India is using these virtual sensors as important analytical components in their test environment. The usage of DataCivet for this application has reduced the GTRE GTX-35VS Kaveri’s development time by several orders of magnitude while guaranteeing that the Machine Learning-based system meets our industrial partner's stringent operating requirements.

Method

Artificial intelligence can generate clever expert systems that lead to company success with the correct data. As a result, we use DataCivet, a solution that pre-processes data with little or no human intervention so that it may be used for further projections and deployment of outcomes.
DataCivet automates the entire data science life cycle, from data preparation to noise detection and removal, high pass filtration of signals and feature extraction, model creation, and final deployment. This allows the organisation to focus on the outcomes of AI/ML applications rather than process issues. Users can enter sensor data into this cloud-based machine learning platform to train an ML model. By executing systematic procedures on raw sensor data and picking models that draw the important information from sensor data, it makes it easier to construct and use ML models in the actual world. The ML model that is produced captures the required trend, resulting in accurate predictions. Using the data characterizations of the failed sensor and its interdependencies with other sensors in the engine, we try to properly anticipate the values of the sensors that have failed.
The model must be fed all of the data points. The training is carried out on several sensors, and it can be seen that the model can accurately detect the trend. The best model is then utilised to create self-learning systems, which can even supplement the sensor or data collecting model by giving correct dummy data.

Result

Despite the fact that various analytical and experimental models of gas turbine engines have been examined and produced to date in order to fully comprehend the complexity of the engines' nonlinear nature, scientists and research professionals are still motivated to continue their work in this area. As a result, for a variety of purposes and applications with higher accuracies and reliabilities, a deeper understanding of several other conventional machine learning techniques is required.

Conclusion

It is obvious that if a solid, consistent data repository is maintained, human resources and time can be saved for future knowledge research. Using patterns identified from the data, the sequence of processes involved in this approach will allow us to generate predictions about the performance of different sections of the GTRE GTX-35VS Kaveri. With a greater understanding of the data and numerous stages of training, we can arrive at a machine learning model that is more accurate. Virtual sensors that can continually monitor the operation of Aero jet engines, even in the presence of deployed sensor failures, have been created using machine learning approaches.Unlike past techniques, the proposed general methodology will be a completely automated data-driven strategy embracing DataCivet that will support the entire data life cycle from data import to virtual sensor production at its end-state. DataCivet was created to allow people with limited technological knowledge and minimal previous statistical experience to build a Machine Learning model like GTRE GTX-35VS Kaveri that fits the needs of a customer, whether automating the entire process to make it more efficient, scaling it up, getting valuable insights, increasing productivity, or reducing expenditure.