Lightweight Self-Detection and Self-Calibration Strategy for MEMS Gas Sensor Arrays

Bing Liu, Yanzhen Zhou, Hongshuo Fu, Ping Fu, Lei FengSensors 2022 · DOI

TL;DR

This paper introduces a lightweight strategy for MEMS gas sensor arrays to automatically detect, isolate, and correct faults in real-time. This is crucial for reliable gas sensing in resource-constrained IoT devices, achieving high accuracy with minimal computational and storage overhead.

What Problem Does This Paper Address?

Intelligent gas sensing systems, especially those utilizing MEMS gas sensor arrays, are increasingly vital for IoT applications. However, these sensors are prone to drift and failure due to environmental factors and material aging, leading to inaccurate readings. Traditional fault detection and calibration methods are often too computationally intensive or require too much memory for deployment on resource-limited edge devices. This paper addresses the need for an efficient, lightweight solution that can ensure the reliability and accuracy of MEMS gas sensor arrays in real-world, constrained environments.

Methods

The proposed strategy consists of two main phases: Self-Detection and Self-Calibration.

  • Self-Detection: This phase uses a Principal Component Analysis (PCA) based model with Squared Prediction Error (SPE) statistics. Before building the model, data preprocessing is performed using Z-Score for outlier removal and Savitzky-Golay filtering for noise reduction. This ensures the robustness of the detection model.
  • Self-Calibration: This phase handles detected faults and drift.
  • Fault Isolation and Recovery (FIR): For single faults, iterative data reconstruction based on SPE is employed. For multiple faults, a Confidence Interval Forecast using Bootstrap (CIFB) method is used, which predicts a confidence interval for sensor readings and identifies outliers.
  • Dynamic Drift Compensation: A dynamic baseline matrix is updated in real-time to compensate for sensor drift, ensuring accurate sensor response over time.

Datasets

The paper utilizes a real-world dataset collected from a mixed gas environment for diabetes exhalation diagnosis. This dataset comprises sensor response data from four MEMS gas sensors (GM-512B, GM-802B, GM-302B, GM-502B) under varying acetone concentrations (0.2 to 5.0 ppm). Faulty testing samples were synthetically generated by superimposing impulse, bias, constant output, and broken-circuit faults onto the normal data.

Key Insights and Ideas

  • Lightweight is Key for IoT: The core insight is that for practical deployment on resource-constrained IoT devices, algorithms must be significantly lighter than traditional machine learning approaches.
  • Multi-Stage Approach for Robustness: Combining data preprocessing (Z-Score, Savitzky-Golay), PCA-SPE for detection, iterative reconstruction for single faults, and CIFB for multiple faults creates a robust system that handles various failure modes.
  • Dynamic Drift Compensation is Crucial: Simply compensating for baseline drift with static methods is insufficient in dynamic environments. A dynamic update mechanism is essential for long-term accuracy.
  • Board-Level Verification Matters: Demonstrating the algorithm's performance on a real embedded platform (CC1350) is critical to prove its practical viability and low resource footprint.

Key Results

  • High Accuracy: The proposed strategy achieved high performance in fault detection, isolation, and data recovery, with detection rates for impulse, bias, constant, and broken-circuit faults reaching 99.3%, 99.6%, 98.0%, and 100.0% respectively, with zero false positives.
  • Computational Efficiency: On the CC1350 platform, the average computation time for the algorithm was a mere 0.28 ms, significantly outperforming ANN (75.03 ms) and RVM (9.25 ms).
  • Low Resource Occupation: The algorithm occupied only 7.89% of the available FLASH memory, compared to 80.85% for ANN and 48.93% for RVM, leaving ample space for other applications.
  • Low Power Consumption: The running power consumption was 9.884 mW, demonstrating its suitability for battery-powered IoT devices.

Limitations and Future Work

  • Limitations: While the paper demonstrates excellent performance, the synthetic generation of faulty data might not perfectly capture all real-world fault complexities. The paper also notes that the probability of single faults is assumed to be much higher than multiple faults, which influences the order of FIR processing.
  • Future Work: The authors plan to focus on applying adaptive correction methods to further enhance the performance of drift compensation.

Summary updated

Abstract

With the development of Internet of Things (IoT) and edge computing technology, gas sensor arrays based on Micro-Electro-Mechanical System (MEMS) fabrication technique have broad application prospects in intelligent integrated systems, portable devices, and other fields. In such complex scenarios, the normal operation of a gas sensing system depends heavily on the accuracy of the sensor output. Therefore, a lightweight Self-Detection and Self-Calibration strategy for MEMS gas sensor arrays is proposed in this paper to monitor the working status of sensor arrays and correct the abnormal data in real time. Evaluations on real-world datasets indicate that the strategy has high performance of fault detection, isolation, and data recovery. Furthermore, our method has low computation complexity and low storage resource occupation. The board-level verification on CC1350 shows that the average calculation time and running power consumption of the algorithm are 0.28 ms and 9.884 mW. The proposed strategy can be deployed on most resource-limited IoT devices to improve the reliability of gas sensing systems.
BibTeX
@inproceedings{liu2022lightweight,
  author = {Bing Liu and Yanzhen Zhou and Hongshuo Fu and Ping Fu and Lei Feng},
  title = {{Lightweight Self-Detection and Self-Calibration Strategy for MEMS Gas Sensor Arrays}},
  booktitle = {Sensors},
  year = {2022},
  doi = {10.3390/s22124315},
}
Figure