Continuous Acoustic Monitoring of Electrical Machines; Processing Signals from USB Microphone & Mobile Smartphone Sensors Detecting DC Motor Controller Fault Jarek Grebenik * , Chris Bingham and Saket Srivastava * The University of Lincoln, School of Engineering, Brayford Pool, Lincoln, LN6 7TS, UK e-mail: {jgrebenik, cbingham, ssrivastava}@lincoln.ac.uk Abstract— Transient current instability is one of the most common faults evident in Pulse Width Modulation (PWM) controlled brushless DC motors. This paper explores the under- developed field of real-time acoustic diagnostics for electrically based faults using consumer grade sensors. Current instabilities produce an audible torque transient on the motor, easily detectable using consumer acoustic sensors; a USB microphone and smartphone in this case. Two time-frequency signal processing techniques, Wavelet Packet Transform (WPT) and Empirical Mode Decomposition (EMD), are used to isolate information pertaining to the fault and are assessed for computational performance. This gives four processed signals to search for instabilities using a peak finding technique. We then compare the performance of each method. With the USB microphone WPT signal correlating the best results (%), a simplistic logarithmic predictive model is used to estimate the durations for the next experimental run, in real-time. The results prove that readily accessible and affordable consumer acoustic sensors can be used for real-time fault diagnostics with a high degree of accuracy. Keywords: acoustic; electric; electrical; fault; detection; diagnosis; smartphone; consumer; microphone; motor; real-time; online; signal processing; wavelet packet transform; WPT; empirical mode decomposition; EMD; time-frequency analysis Abbreviations: AE – Acoustic Emission, EEMD – Ensemble Empirical Mode Decomposition, EMD – Empirical Mode Decomposition, IMF – Intrinsic Mode Function, PWM – Pulse Width Modulation, WPT – Wavelet Packet Transform I. INTRODUCTION Acoustic supervision and monitoring is an attractive prospect for many industrial applications with proven advantages over many established systems including; earlier and more accurate detection, non-invasive and readily accessible sensors and better cost-benefit performance [1]. Research in this field is highly active with the majority focussed on seeded mechanical impact based faults, detecting the acoustic shockwave. Acoustic Emission (AE) energy indexing/analysis has been used to detect seeded mechanical defects on roller element bearings [2, 3]. Similar case study is also explored in [4, 5, 6] using wavelet analysis. A good example of vibration analysis for bearing fault detection using wavelet transform is [7]. Another study investigates seeded mechanical defects on roller element bearings and gears using EMD of acoustic signals [8]. The work in [9] uses wavelet signal processing of vibration and acoustic signals to detect a simulated cracked tooth in a gear box. Another study mentions the requirement for real-time condition monitoring in industry and looks at motor fault diagnosis. However, it is unclear whether the experiment demonstrates real-time detection, and this is an area that is highly underdeveloped in this field [10]. All prior works discussed use specialised, research grade AE sensors, not easily accessible, that require mounting to the component to deliver sufficient signal to noise ratio. A review of condition monitoring and fault diagnostics for electrical motors is given in [11]. However, all the faults in this work are seeded and there is little mention of acoustic techniques. There is very little literature on acoustic diagnostics for electrically-based faults nor for real-time processing. As opposed to seeded mechanical faults, it is far more difficult to detect and model transient current instabilities arising in PWM-based DC motors. Electronically commutated motors require a closed-loop power controller to convert the DC supply to AC for each phase; synchronising the motor. The design and operation of each controller will vary depending on the supply, type of motor and application. Often, controllers under commissioning, motor tuning, or operating in widely varying conditions, can result in transient current instabilities delivered to the motor. These instabilities could potentially cause damage to the controller or demagnetisation of the motor [12]. Detection could allow intervention to prevent further damage, and diagnostics could help assess maintenance requirements. We present a novel application of acoustic monitoring to detect electro-mechanical instabilities on a brushless DC electric motor. Our work explores aspects previously unaddressed in the literature, namely; real-time processing and expansion to include electrically based faults. Consumer sensors (which are easily accessible) are used to highlight the under-usage of these types of sensor in both research and industry. Acoustic signals are processed in real-time to reveal the instability time-frequency information. This is done using two signal processing techniques, WPT and EMD, which are analysed for computational performance and ability to deliver information pertaining to the fault. Our proposed solution uses a bespoke PWM power control system that delivers intermittent transient current instabilities with increasing frequency for higher voltages. At the end of the previous PWM cycle, the controller calculates the new duty from the demand. When the sampled current is already too high, the next cycle is switched off. This can resolve or continue to the end of the commutation period. A detailed view of the instability is given in Figure 1. This fault is native to this particular setup and is used here as an example of the potential electro-mechanical faults requiring diagnostic information. Acoustics are well suited to this application due to the audible torque transient caused by the current instabilities; hence an electro-mechanical fault. In this case, the frequency