An Embedded System for Real Time Vibration Analysis Giuseppe Merendino( * ), Augusto Pieracci( # ) , Massimo Lanzoni( # ) and Bruno Riccò( # ), Fellow IEEE ( * ) T3lab Via Bassanelli 9/11– 40129 Bologna giuseppe.merendino@t3lab.it ( # ) DEIS University of Bologna V.le Risorgimento 2 – 40136 Bologna Abstract This paper presents a new μController-based embedded system devoted to vibration analysis for fault diagnosis in rotating machine. The developed system is based on advanced spectra methods, such as Wavelet and Fourier Transform, to emphasize the harmonic content of the considered signal thus facilitating early fault diagnosis. Suitable algorithms have been developed for such complex methods to run on a μC featuring limited computing resources. The cost and performance of our system are compatible with integration into industrial machines for continuous monitoring of their status. Moreover, the developed system is easily customizable and adaptable to a variety of automatic industrial machines. Introduction Effective and continuous monitoring of industrial machines by means of electronic systems “embedded” in their structure allows real time assessment of the machine status as well as optimized maintenance, i.e. replacement of really damaged parts instead of unconditional substitution at predefined time. To this purpose, analysis of vibration signals picked-up in sensitive parts of the machines by means of suitable sensors represents an established technique, in particular for application to rotating machines, and powerful algorithms have been developed to extract from the vibration signals the information needed for effective monitoring and diagnosis. Unfortunately, such algorithms need intensive calculations, hence normally require expensive systems based on industrial PCs. Instead, detailed monitoring would benefit from an “embedded” solution, featuring networks of sensor nodes distributed throughout the machine (and possibly in wireless communication with one another as well as with a centralized monitor). This type of solution, however, requires adapting complex signal processing algorithms to the limited computer resources of low-cost micro-Controller (μC). In this context, the present work describes a new μC-based embedded system devoted to vibration analysis for fault diagnosis in rotating machine, that implements advanced methods for signal analysis, such Fast Fourier Transformations (FFT) and Discrete Wavelet Transformations (DWT), suitably adapted to run on a low-cost μC (32 bit RISC ARM9 STR912F44W). Such e system makes use of micro- fabricated accelerometers as sensors. Experimental results obtained with a Lab model of a rotating machine, allowing easy emulation of common faults, clearly indicate that the developed system features good diagnostic capabilities, in that different types of faults can be distinguished and recognized. The performance of the system have also been satisfactorily tested by means of artificially synthesized signals. State of the Art For the applications of interest here, vibration signals can be analyzed using time domain averaging, a heavy processing technique to extract periodic waveforms from noisy signals that, however, requires either signal pre-processing or sampling synchronization performed with ad-hoc hardware. Furthermore, time domain analysis becomes difficult in the presence of multi-tone noisy signals. On the other hand, signal analysis in the frequency domain is essential for diagnostics, since different types of faults produce specific deformations in the spectrum harmonic components. For these reasons, Halim et al. [1] proposed the Continuous Wavelet Transform (CWT) as an approach combining the advantages of both time and frequency domain analysis. Unfortunately, such an approach is very demanding in terms of computing resources, hence direct implementation on μCs (and even DSP) is impossible. In particular, [1] presents a technique based on time- frequency domain averaging that combines Continuous Wavelet and time domain averaging. A test rig is proposed for data generation and used to assess the effectiveness of the proposed technique, but no implementation in the form of an embedded system is attempted. [2], instead, proposes a technique for vibration analysis in motors based on FFT and Discrete Wavelet Decomposition, but a Digital Storage Oscilloscope is used for data collection and processing. [3], on the other hand, introduces a system based on the FFT for continuous monitoring of machines’ status, but the implementation makes use of a PC-Board with two DSP’s . In [4] a FPGA based vibration analyzer for machine monitoring is presented that uses (only) the FFT and experimental results are provided only for unbalanced or damaged motors. [5] makes use of Discrete Wavelet Analysis to study the effect of damage only in ball bearing. An experimental set-up is realized to generate signals that are sampled and processed by means of a PC. Finally, [6] proposes a system where a μC is responsible (only) for data acquisition and communication with a PC, where all signal analysis is performed. 6 978-1-4577-0624-0/11/$26.00 ©2011 IEEE