TECHNOLOGY AREA(S): Materials
OBJECTIVE: Address issues pertaining to machine health through Monitoring and Diagnosis via Electrical Waveform Auditing
DESCRIPTION: Some types of machine health threats (fault or attack) may be subtle and not necessarily change the power consumption of machines, but cause distorted electrical waveforms (e.g., increased harmonics) in power networks which may affect the precision and functions of electrical machines. The attacks may be direct or indirect. In direct attacks, a malicious device may be plugged into a power outlet and inject harmonics to the power network of machines. The indirect methods may involve hacking and controlling an electrical machine in order to generate distorted electrical waveforms to affect other machines in the power network. For instance, if denial of service (DOS) occurs, the system might become unstable, resulting in unusual harmonics and torque ripples, which later affect product quality. In this task, it is proposed to analyze electrical waveform data from strategically-placed sensors in manufacturing systems for health monitoring and diagnosis. No record in industry suggests this approach has not been attempted before. To achieve this goal, a high dimensional analysis method and information incorporation mode would need to be developed. Traditional time series analysis or machine learning methods ignore some unique characteristics of the multi-stream measurement data; in particular, the coexistence of strong temporal correlation and inter-stream relatedness is not accounted for. The machine learning formulation proposed in this task for multiple time series is intuitively nonparametric regression in statistical learning theory, which uses multiple coevolving time series data to capture both the temporal dependence and inter-series relatedness.
PHASE I: Build the relationship model between system statistics (e.g. “normal state, controller attack, attack, DOS attack, short circuit fault,” etc.) and electrical waveform data (e.g, total harmonic distortion, current ripples, voltage/current unbalance, etc.). Firstly, validation that electrical waveform of manufacturing systems can be used to detect cyber and physical attacks would need to take place. Then, a disaggregation model to map the relationship to assist root cause diagnosis could be developed.
PHASE II: Develop monitoring and diagnosis software to classify the observed data into “trend” functions and anomalies based on the “normal” behavior data and simulated “faulty” data. Once the trend and fault libraries are built, the monitoring system detects anomalies when the fitting error is larger than the threshold. Then a classification model can be learned to classify the threat source to the most possible location.
PHASE III: Monitoring and diagnosis via electrical waveform have many commercial applications. A successful system could be marketed to commercial manufacturing, aerospace industry as well as other defense customers. Additional markets might include the smart home, construction, and power industries.
REFERENCES:1. F. Li, B. Yang, J. Ye, and W. Song, “Generator fault diagnosis based on sparsely placed sensors in power networks,” Sensors, 2019, submitted.; 2. B. Yang, F. Li, J. Ye, and W. Song, Condition Monitoring and Fault Diagnosis of Generators in Power Networks Conference IEEE Power & Energy Society General Meeting, 2019.; 3. J. Guo, J. Ye, and A. Emadi, “DC-Link current and voltage ripple analysis considering anti-parallel diode reverse recovery in voltage source inverters,” IEEE Transactions on Power Electronics, vol. 33, no. 6, pp. 5171-5180, June 2018.; 4. F. Peng, J. Ye, A. Emadi, and Y. Huang, “Position sensorless control of switched reluctance motor drives based on numerical method,” IEEE Transactions on Industry Applications, vol. 53, no.3, pp. 2159-2168, May-June 2017.
KEYWORDS: Manufacturing System, Monitoring And Diagnosis, Electrical Waveform