Statistical Time Energy Based Damage Detection in Steel Plates Using Artificial Neural Networks Paulraj M P, Mohd Shukry Abdul Majid, Sazali Yaacob, Mohd Hafiz Fazalul Rahiman and R.Pranesh Krishnan School of Mechatronics Engineering, Universiti Malaysia Perlis, Perlis, Malaysia. praneshkrishnan@gmail.com Abstract- In this paper, a simple method for crack identification in steel plates based on statistical time energy is presented. A simple experimental procedure is also proposed to measure the vibration at different positions of a steel plate. The plate is excited by an impulse signal and made to vibrate; statistical features are then extracted from the vibration signals which are measured at different locations. These features are then used to develop a neural network model. A simple neural network model trained by back propagation algorithm is then developed based on the statistical time energy features to classify the damage location in a steel plate. The effectiveness of the system is validated through simulation. Keywords- Time domain, Damage Detection, Back Propagation neural network. I. INTRODUCTION Health monitoring of vibrating structure in machines is a important task in industries. Damages can put human safety at risk, cause long term machine downtimes, interruption in the production and subsequently increase the production cost. Early damage detection and possible location of the faults from the vibration measurements is one of the primary task of condition monitoring. Condition monitoring enables early detection of faults. In recent years there has been an increasing interest in the development of online condition monitoring systems due to the success in several applications. A damage condition of a steel plate can be detected by the vibration signal propagating through it, when it is subjected to an impulse force. There are many technologies that have been developed to detect the faults in a gear box, bridge structures and bearings. The existence of a crack in a steel plate reduces the stiffness of the plate and this reduction in stiffness ultimately reduces the natural frequencies. Further, this also changes the mode shape of vibration. An analysis of the propagation of the vibration signal makes it possible to detect the fault. An extensive literature review of the state of art of vibration analysis and damage detection has been recently published by S.W. Doebling [1]. A detailed survey of the state of art in the damage detection field using modal analysis has been presented by Richardson[2[. A detailed review of the different vibration and acoustic methods such as the time and frequency domains, acoustic emission technique are presented by Tandon and Nakra [3]. Using fracture mechanics method, Dimarogonas [4] and Anifactis [4] computed the equivalent stiffness and developed a model for crack detection in beams. An experimental technique to estimate the location and depth of a crack in a beam has been developed by Adams and Cawley [6]. The methodology of crack detection based on natural frequency changes has been closely studied by Shen and Pierre [7]. In this paper, it is proposed to detect a faulty location based on the statistical time energy based features extracted from the vibration signal. II. EXPERIMENTAL DESIGN AND DATA AQUISITION A. Data Acquisition System (DAQ) Measurements of the vibration signals are acquired using a LMS SCADAS Mobile SCM01 Data Acquisition System. This system has 4 input channels and Ethernet connectivity to the LMS Test Lab 8A software. The features supported by the LMS SCADAS are: a maximum sampling frequency range of up to 102.4 kHz per channel, 105 dB signal to noise ratio and a high speed Ethernet connection. The DAQ system is monitored through the LMS Test Lab software which supports a wide range of applications. . B. Vibration and Pressure Transducers Accelerometers are Vibration transducers which possess high natural frequencies compared to the vibration to be measured and indicate acceleration [8]. The piezoelectric accelerometers are widely preferred over the digital accelerometers in many applications due to its high accuracy and sensitivity. The general purpose Piezoelectric accelerometer with an input sensitivity of 10 / 31.6 / 100mV/g (g = 9.82 m/s 2 ) and a resonant frequency of 28 kHz is used in this experimental work. Force transducers are used to produce impulse forces and commonly used for impact tests. The general purpose force transducers or so called impact hammer (Dytran 5800B2 - 50LbF range, 100 mV/LbF) is used in this research work. C. Experimental Setup A simple experimental design to test the structure in a simply supported condition is proposed in this paper. An aluminum test rig of dimensions (90x60x3) cm is fabricated 978-1-4244-4152-5/09/$25.00 ©2009 IEEE 34 2009 5th International Colloquium on Signal Processing & Its Applications (CSPA)