Center for Intelligent Robotics
實驗室名稱：機電系統整合實驗室
研究專長：人工智慧、訊號分析、故障診斷、自動化、機器人及控制
Name of Lab：Electromechanical System Integration Lab (ESILab)
Research Interest：Artificial Intelligence, Signal Analysis, Fault Diagnosis, Automation, Robotics and Control
電動機故障診斷馬達故障分析
Electric Motor Fault Diagnosis
振動分析
-
機械故障引起的振動異常的早期檢測和分類
-
研究方法包括時域和頻域信號分析、多分辨率分析和回歸模型估計
-
出色的故障診斷和識別方法
電信號分析
-
早期檢測和分類電氣和機械故障
-
研究方法包括基於信號的分析和回歸模型估計
-
最簡單的故障診斷和識別方法
Vibration Analysis
-
Early detection and classification of the vibration anomalies caused by mechanical faults
-
Research methods include time-domain and frequency-domain signal analysis, multiresolution analysis, and regression model estimation
-
Excellent approach for fault diagnosis and identification
Electrical Signature Analysis
-
Early detection and classification of both electrical and mechanical faults
-
Research methods include signal based analysis and regression model estimation
-
Most simple approach for fault diagnosis and identification
時頻信號分析
Time-Frequency Signal Analysis
由於使用了靜止正弦曲線，基於傅里葉的分析方法誤解了非線性動力學的真實物理學。無法準確識別高節距刀具顫振、高速加工時工件與刀具之間的異常振動等動態加工不穩定性，從而因刀具過早失效、表面光潔度不均勻和生產率低下而造成巨大損失。為了處理到混沌的獨特特性，需要同時捕捉時間和頻率的特徵。因此採用希爾伯特-黃變換（HHT）來提供更準確的系統特性和非線性的瞬時幅度和頻率。在本研究中，HHT 用於分析來自力傳感器、加速度計、機器加工過程中的光學傳感器和麥克風。研究路線圖見下圖：
Fourier-based methods of analysis misinterpret the genuine physics of nonlinear dynamics due to the employment of stationary sinusoids. They fall short in precisely identifying dynamic machining instability such as high-pitch tool chatter, the aberrational vibration between the workpiece and cutting tool when machining at high speed, henceforth causing huge losses due to premature tool failure, uneven surface finish, and low productivity. To deal with the unique characteristics of route-to-chaos, it is necessary to capture the features of time and frequency simultaneously. Hence Hilbert-Huang transform (HHT) is adopted to provide more accurate instant amplitudes and frequencies of system characteristics and nonlinearities. In this research, HHT is used to analyze the measurement from force sensor, accelerometer, optical sensor and microphone during machines process. The research road map can be seen in the following figure:
超音波加工
這是一個通過我們實驗室開發的 HHT 程序分析超聲波混合器的軸向振動的例子。下圖顯示了光學傳感器測量的時間響應。振動幅度約為 40 微米，採樣率為 52,000 Hz。
Ultrasonic Machining
This is an example to analyze the axial vibration of an ultrasonic mixer by the HHT program developed in our lab. The figure below shows the time response of the measurement from an optical sensor. The vibration amplitude is about 40 micron with a sampling rate of 52,000 Hz.
左圖顯示了超聲波加工過程的傅里葉譜。它僅識別 20,000 Hz 處的共振頻率，但掩蓋了共振以外的頻率。由於傅里葉變換的特性，它無法研究特定時間實例的頻率變化。右圖顯示了通過本徵模式分解 (EMD) 和希爾伯特變換獲得的瞬時頻率。結果表明，有六個頻率範圍從 500 Hz 到 20,000 Hz，並且可以觀察到它們隨時間的變化。
The left figure below shows the Fourier spectrum of the ultrasonic machining process. It only identifies the resonance frequency at 20,000 Hz but obscures the frequencies other than resonance. Due to the characteristic of Fourier transform, it is unable to investigate the frequency change at certain time instance. The figure on the right shows the instantaneous frequency obtained through Intrinsic Mode Decomposition (EMD) along with Hilbert transform. It is shown that there are six frequencies ranged from 500 Hz to 20,000 Hz and their variation with time is observed.
機器人力回饋控制
Robot Force Control
傳統的阻抗控制使用二階微分方程來解決機器人手臂與環境之間的相互作用。應用質量（M）、剛度（K）和阻尼（B）係數的合理估計。然而，機器人手腕處的力測量不僅包括環境施加的力，還包括重力和科里奧利力。並且質量、剛度和阻尼係數隨時間而變化。有必要改進傳統的阻抗控制結構，如下所示。
The conventional impedance control uses a second order differential equation to address the interaction between the robot arm and the environment. Reasonable estimation of mass(M), stiffness(K) and damping(B) coefficients is applied. However, the force measurement at the robot wrist not only includes the force exerted from the environment but gravity and Coriolis force. And the mass, stiffness and damping coefficients are subject to change with time. It is necessary to improve the conventional impedance control structure as shown below.
非線性時頻控制
Nonlinear Time-Frequency Control
為了處理到混沌的獨特特性，需要同時捕捉時間和頻率的特徵。傳統控制理論無法控制非線性系統的原因是因為它們假設靜態系統動力學，並且它們不能控制時域和頻域上的同時惡化，這是通往混沌的特徵。在研究的基礎上，提出了一種新的非線性控制理論，以解決和保留所有經歷到混沌的非線性系統所固有的基本特徵。一個不需要線性化或封閉形式的解決方案，因此它沒有所有先前方法的缺點並保留了系統的真正底層特徵。它同時控制時域和頻域，而不會扭曲或曲解真實動態。所提出的控制理論對包括精密製造和國家安全在內的廣泛應用產生了巨大影響。它抑制了高速微銑削加工過程的衰減不穩定性，該過程經常使當前的非線性控制器失效，並被證明是一種通用的通信解碼器。
To deal with the unique characteristics of route-to-chaos, it is necessary to capture the features of time and frequency simultaneously. The reason that conventional control theories fail to control nonlinear system is because they assume static system dynamics, and they don’t control the simultaneous deterioration on time and frequency domain, a signature of route-to-chaos. Based on the investigations, a novel nonlinear control theory is formulated to address and retain the fundamental characteristic inherent of all nonlinear systems undergoing route-to-chaos. One requires no linearization or closed form solution so that it doesn’t have drawbacks of all previous methods and preserves the genuine underlying features of the system. It controls in the time and frequency domain simultaneously without distorting or misinterpreting the true dynamics. The proposed control theory has huge impact on a broad range of applications including precision manufacturing and national security. It restrains the decaying instability of high-speed micro-milling process which often fails the current nonlinear controller, and it is proofed to be a universal decipher in communication.
非線性電路混沌同步以下是同步兩個具有不同初始條件和驅動頻率的非自治混沌電路的示例。同步過程被高頻噪聲破壞。結果表明，即使存在噪聲，x、y 和 z 中的時域誤差仍被充分限制在稱為實際同步的有限範圍內。並且響應電路的頻率恢復到與驅動信號的帶寬相同，但特性並不完全相同。
Chaos Synchronization of Nonlinear Circuits The following is an example to synchronize two nonautonomous chaotic circuits with different initial conditions and driving frequencies. The synchronization procedure is corrupted with high frequency noise. It is shown that even with the noise, the time domain errors in x, y, and z remain adequately constrained within limited range called practical synchronization. And the frequencies of the response circuit are restored to be of the same bandwidth of the driving signal, though not of exactly the same characteristics.
