THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS TECHNICAL REPORT OF IEICE. E-mail: {ytamura,takai,tkato,tm}@vision.kuee.kyoto-u.ac.jp Abstract Current Wave Pattern Analysis for Anomaly Detection of Electrical Devices Yuto TAMURA, Takeshi TAKAI, Takekazu KATO, and Takashi MATSUYAMA Graduate School of Informatics, Kyoto University E-mail: {ytamura,takai,tkato,tm}@vision.kuee.kyoto-u.ac.jp This paper proposes a novel method to detect and evaluate anomaly of an electric appliance from variations of current waveform patterns that are measured under ordinary usage. Since the electrical appliance is regarded as a large electrical circuit, a specific pattern of the current waveform can be observed from one appliance and another. This indicates that anomaly or degradation can be detected by observing a variation of the current waveform patterns from those of the normal condition. The appliance, however, autonomously controls itself depending on its internal states and an external environment, and then the waveform changes its pattern every moment. Thus, variations caused by those controls or anomaly should be distinguished from each other. In this paper, we propose a method to detect anomaly or degradation by analyzing variations of the current waveform obtained from each control state, which is identified from power consumption patterns of the appliance. The experimental results using a fan and a refrigerator show that we can detect variations of the current waveform patterns caused by anomaly or degradation and evaluate them both qualitatively and quantitatively. Key words Anomaly detection, Degration detection, Current waveform pattern analysis, Power data segmentation.
[] []........ 3. 4. [3] khz 337.
Power[W] 3 4 time[min] Scale-space filtering (SSF) [6] 3 a ) Scale-space filtering (SSF), 3 F (z, σ) = f(z) g(z, σ) () g(z, σ) = F (z, σ) z =, b ) z exp{ πσ σ } () 3 F (z, σ) z 3 = (3) 3 root mean-square errors: RMSE 4 RMSE 5 3(a) c ) RMSE 3(b) RMSE 5 5 5 3 5 5 5 3 time [min] (a) 3 5 5 5 3 5 5 5 3 time [min] (b). 3. 4 3
m k b k I n (= [i n,..., i n337 ]) 4 a nk k k a nk = (I n m) b k (4) 3 m Σ I n D n (5) D n = (I n m) T Σ (I n m) (5) D = N D N n= n N D 3. 3. 3 4 44.7W 43.9W 45.5W 3.9W 4 5 5 5 5 time[min] 5 9% 5(a) 5(b).79.75 5(a) 6 3% 4
.6 Normal Overturn.4 ForcedStop...4.6.5.5 (a) 5 3 3 (b) Normal Racing 9% 6. 4.7 8.9867 6. 8. 4..455.379. 5 5 y y 4y 6y 3y 5 5 3. 7 4 6 3 8 9% 9 4 3.5 3.9 3.49.8.5.988.75 Normal y 4y 6y 3y 7 9 3.. 3...749,.8886,.95,.88,.743 5
state state state3 3 4 state state state3 3 4 state state state3 3 4 state state state3 3 4 state state state3 3 4 time[min] state, state,..., state 5.5.5.5.5.5.5 3 3 (a) state 3 3 (d) state 4 y y 4y 6y 3y y y 4y 6y 3y.5.5.5.5.5.5 3 (b) state y y 4y 6y 3y y y 4y 6y 3y 3 3 3 (e) state 5 6 4 4 y y 4y 6y 3y 6 6 4 4 6 (c) state 3 9% State State State 3 State 4 State 5 8 4. 4 3.5.5.75. 3.347 3.9 3.5 3.9 3.57 3.99.43.34.46.564 Normal y 4y 6y 3y 3.83 3.467 3.434 3.7 3.8 3.85.8.899.873.83 [4], [5] [] [] Gaussian Processes PRMU 75 MVE, pp. 6,. [3], i-energy Profile:, B, Vol.J94-B, No., pp. 3 45,. [4] Interval-based switching Kalman filters, USN 4, pp. 39 44,. [5] ASN3 3 pp.59 6 3. [6] Witkin, A. P. Scale-space filtering, Proc. 8th Int. Joint Conf. Art. Intell., Karlsruhe, Germany, pp. 9 -, 983. 6