Enhancement of MEMS-based ultrasonic transducers for sensing applications using machine learning

Abstract

This thesis presents an in-depth exploration of the integration of ultrasonic signal transmission, ultrasonic signal processing, and advanced fault detection methodologies, highlighting the synergy among these technologies through three interrelated studies. The first study introduces a high-precision airborne ultrasonic rangefinder system based on piezoelectric micromachined ultrasonic transducers (PMUTs). Utilizing both transmitters (Tx) and receivers (Rx) fabricated with highly sensitive piezoelectric beams, the system operates on the time-of-flight (ToF) principle, where ultrasonic waves travel between the Tx and Rx to detect obstacles or calculate distances. Enhanced by deep learning models—specifically convolutional neural networks (CNNs) with k-fold cross-validation—the rangefinder achieves superior accuracy compared to traditional methods, ensuring robust performance in diverse environments. This makes it suitable for applications in robotics, augmented reality, and industrial safety systems. The second study focuses on the application of ultrasonic signal processing for industrial fault detection, employing only the receiver (Rx). Concentrating on pipelines and motor systems, this research utilizes the Rx to detect mechanical faults by capturing ultrasonic signals and applying advanced feature extraction techniques in both time and frequency domains. Key features—including skewness, kurtosis and crest factor—are extracted, and dimensionality reduction methods like principal component analysis (PCA) and linear discriminant analysis (LDA) are employed to streamline the data. Machine learning classifiers such as k-nearest neighbors (KNN), support vector machines (SVM) and decision trees (DT) are used in conjunction with ensemble learning techniques like stacking and boosting to enhance accuracy and reliability in fault detection. The system's ability to detect and classify faults in real time is further validated by deploying the models on microcontroller units (MCUs), underscoring its potential for real-world applications. The third study explores the use of ensemble machine learning models for real-time fault detection using ultrasonics. Emphasizing scalability and efficiency, this research demonstrates the use of ultrasonic signals for monitoring industrial systems such as pipelines and rotating machinery. The Rx plays a central role in capturing ultrasonic signals, which are then processed and analyzed by machine learning algorithms to detect faults like leaks, blockages, and bearing failures. The study compares several ensemble learning models—including gradient boosting (GB), voting classifiers, and AdaBoost—and evaluates their performance using k-fold cross-validation. Deployment of these models on resource-constrained devices like ARM Cortex-M4F MCUs demonstrates the feasibility of real-time fault monitoring on embedded systems. This approach not only addresses current industrial needs but also paves the way for future innovations in wireless sensing and smart monitoring systems. In conclusion, the research presented in these studies underscores the promising role of ultrasonics, enhanced by machine learning, in augmenting industrial system efficiency, safety, and functionality. By integrating ultrasonic sensors with advanced machine learning techniques, the developed systems provide robust, real-time fault detection and diagnostics, offering significant improvements over traditional monitoring methods. The successful implementation of these technologies addresses pressing industrial challenges—such as early fault detection in pipelines and machinery—and lays the groundwork for future innovations in wireless sensing, embedded systems, and real-time monitoring across various sectors. This work establishes a strong foundation for ongoing research into more efficient, scalable solutions for industrial automation, predictive maintenance, and safety monitoring.

Date	14 Feb 2025
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Frédéric Nabki (Supervisor)