Designing Fault Tolerant Wireless Biosensor Network with Composite Techniques

Abstract

Reliability is an essential feature of Wireless Bio Sensor Networks (WBSNs) that can be deployed for continuous health monitoring purposes. The major concern of masking caused by the traditional sliding window method utilised for fault diagnosis has been solved with this new approach to fault diagnosis within WBSN networks. The approach presented in this paper provides three ways to address the critical masking associated with sliding window techniques. A Locked Historical Baseline Detection (LHBD) mechanism is employed to prevent the masking of persistent faults by locking calibration statistics during initialization. Secondly a CUSUM (Cumulative Sum) integration method is applied to detect fault drift. Then an ensemble voting technique is projected to improve accuracy in finding faults. An empirical evaluation of the proposed method versus Re-Implementing the sliding window in identical conditions (30 trials × 6 fault probabilities × 3 different types of faults) indicates that the proposed method achieved 98.3% accuracy compared to 88% for the Re-Implemented sliding windows (+11.6%, p < 0.001); Furthermore, persistent faults were no longer masked due to a reduction in false negative rates from 12.7% to 0.03% (approximately 97.6%), whereas persistent faults were masked by the traditional sliding window method after ~30 minutes. Paired t-tests are used to statistically validate each improvement at the 95% confidence level. In clinical WBSN situations, where missing a problem which is of Type II mistake, has more serious repercussions than false alarms, the methodologies especially well-suited.