Do you want to explore the intersection of TinyML, embedded signal processing, and real-time sensor health diagnostics?

Sensors operating in harsh production environments face extreme conditions such as dirt, vibration, and mechanical wear, often leading to subtle sensor drift and degraded data quality long before a complete failure occurs. In this thesis, we invite one or two students to develop and evaluate lightweight anomaly detection algorithms that execute directly on resource-constrained microcontrollers (Edge/TinyML). The goal is to analyze signal health patterns in real time at the sensor level, flagging data quality issues before bad data reaches downstream systems.

About the project

The aim is to design, implement, and evaluate an edge-based self-diagnosing sensor system capable of automated drift and defect detection. Within this scope, you will:

• Work with industrial time-series datasets (such as the NASA Turbofan degradation dataset) and model realistic sensor fault profiles.

• Develop lightweight anomaly detection and signal-processing algorithms in Python and optimize them in C/C++ for microcontroller deployment.

• Implement real-time inference on resource-constrained hardware (e.g., ARM Cortex-M or similar microcontroller platforms).

• Benchmark system performance in terms of detection accuracy, memory footprint, power efficiency, and the ability to distinguish sensor faults from true physical process anomalies.

Possible Research Questions

• RQ1: Which lightweight anomaly detection algorithms provide the best trade-off between detection rate and computational resource efficiency when executing directly on a microcontroller?

• RQ2: How can edge-level algorithmic analysis effectively differentiate between true physical process anomalies and sensor-related fault modes (e.g., signal noise, connection degradation, or calibration drift)?

Features & technologies

• TinyML & Embedded Systems (C/C++, TensorFlow Lite for Microcontrollers, Edge Impulse, Microcontrollers)

• Time-series Anomaly Detection & Signal Processing (Python, PyTorch, Scikit-learn, SciPy)

• Sensor health monitoring, drift analysis, and predictive maintenance

• Industrial time-series datasets (e.g., NASA Turbofan / C-MAPSS)

Who are you?

We are looking for one or two master’s students with an interest in embedded AI, signal processing, and industrial machine learning. Proficiency in Python and C/C++, as well as foundational knowledge in signal processing, statistics, or machine learning, is required. Experience with time-series data, sensor networks, or microcontroller development is a plus. Most importantly, you are curious, motivated, and eager to apply TinyML to real-world hardware challenges.

To give you the best possible support during your thesis, we’d like you to be able to come to the office connected to the project and spend most of your time working from there.

Application:

We look forward to receiving your resume, and preferably, a personal letter in which you explain why you want to write your thesis with Syntronic.

We screen and evaluate applications on an ongoing basis. The thesis project may be filled before the application deadline.