Physically Segmented Hidden Markov Model for Continuous Condition Monitoring
A temporal probabilistic approach of physically segmented hidden Markov model (SHSMM) is developed for continuous condition monitoring. The approach has the advantage of providing an explicit relationship between the actual health states and the hidden state values. SHSMM provides good result in both diagnosis and prognosis.
Contact PersonZhou Junhong(jzhou@SIMTech.a-star.edu.sg)
- Continuous Health Assessment Framework:
A frame work for offline model build up and online machine health assessment
- Feature Extraction and Selection:
Both Statistical Features and wavelet features are extracted. Fisher's Discriminant Ratio is used to find a proper subset of features.
- Hidden Semi-Markov Model (HSMM):
HSMM estimate the probability of remaining in the same state using the duration distribution, and using the transition probability to estimate the proceeding to the next health state.
- Diagnostics and Prognostics:
The developed HSMM model algorithm diagnosis the machine condition and discovery the root causes. The developed HSMM model algorithm also able to predict the future health state at time t (t > T) while the observation data is only available up to the current time
BenefitsThe approach has the advantage of providing an explicit relationship between the actual health states and the hidden state values. SHSMM provide good result in both diagnosis and prognosis. 3 % improvement in precision and about 50% improvement in standard deviation have been achieved for tool wear prediction.
The developed methodology will be used for critical machine condition health assessment of following industries:
- Precision Engineering
- Oil & Gas