AN ARTIFICIAL NEURAL SYSTEM FOR HEALTH MONITORING OF LARGE STRUCTURES Goutham R. Kirikera, Saurabh Datta, Mark J.Schulz Smart Structures and Bio-Nanotechnology Lab Department of Mechanical Engineering University of Cincinnati, Cincinnati, OH 45221-0072 Mannur J. Sundaresan Department of Mechanical Engineering North Carolina A&T State University, Greensboro, NC 27411
Overview Introduction Simulation Experimental Results Detection of Simulated Acoustic Emission Comparison of Experimental and Simulated Results Conclusion
1. INTRODUCTION Currently, Structural Health Monitoring (SHM) is not widely adopted This is because too many individual sensors are required, and sequential signal processing of sensor data is not efficient Highly-distributed interconnected sensors and parallel processing may simplify and make SHM more practical Miniature distributed sensor nodes can be formed using piezoceramic sensors. These nodes (~10) can be connected in series to form a continuous sensor Continuous sensors (~20) in turn can form an Artificial Neural System (ANS) The ANS can measure low frequency dynamic strains and high frequency ency Acoustic Emission (AE) signals to identify damage growth
Wave Simulation Algorithm Features of the CODE Panel Fiber glass composite Sensor PZT or AFC Number of sensors - 100 or more Input Step or Impulse
Equations used in the Simulation Average strain Equation in the x-direction is h 1 mπ nπ ε x ( t) = ΣΣ( / ) a 2 x y n m a b Where x x = x 2 x 1 and y y = y 2 y 1 h = Plate Thickness m,n = Modes a = Length of the plate b = Width of the plate mn mπx ( t)(cos a 2 mπx cos a 1 nπy )(cos b 2 nπy cos b 1 )
Simulation of a 10 by 10 Neural System Plate = Fiber-Glass composite Dimensions = 48 x 48 Thickness = ¼ Total Number of Sensors = 100 Number of Neurons = 20 (10 Rows + 10 Columns) Lead Break = Simulated by Acoustic Emission Sensor = PZT
Channel 1 (columns output) Channel 2 (rows output) An ANS in a composite panel in which 100 bi-directional AFS nodes (N1-N100) form 20 neurons with outputs (V1-V20) where the final output of the neural system consists of 20 on/off neuron firing signals and only two time history signals
Response of the Plate due to an Impulse Force Time = 2.49 e-4
Response of the plate due to an Impulse Force Time = 7.51e-4 sec
Response of the Plate due to an Impulse Force Time = 13 e-4
ANS Strategy Develop the ANS using smart materials, microelectronics, and new signal processing Programmable Signal Transducer (PST) a. Series/Parallel neurons, b. Channel switching, c. Signal acquisition, d. Low/high freq. filtering, e. Sensor self-check Computer Signal Processing Diagnostics Prognostics The ANS will replicate the efficiency of the Biological Neural System to sense dynamic strain as pain. Dual node sensors are used for this example Signals from neurons
Voltages from the Neurons (Y- Direction only) 0.15 0.1 Voltage (Volts) 0.05 0-0.05-0.1 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 Tim e x 10-3
10 Firing of the Neuron S witches in the Y -Direction 9 8 7 Neuron Number 6 5 4 3 2 1 0 0 0.2 0.4 0.6 0.8 1 Tim e (S ec) 1.2 1.4 1.6 1.8 2 x 10-3 Neurons 5 and 6 fire together indicating that the AE is initiated between them.
1.4 Individual Responses of the Y-Direction Neurons 1.2 1 Voltage (volts) 0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 Tim e (S ec) x 10-3
2.5 Com bined Output of the Y-Direction Neurons System O utput Voltage(Colum n) 2 Voltage (volts) 1.5 1 0.5 0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 Tim e (S ec) x 10-3 The system output in x and y directions are the same because of the symmetry in the location of sensors.
Summary Only 2 channels of data acquisition and 20 on/off switches are required. The simulated Acoustic Emission can be located in a 4 X4 area on a 48 X 48 square plate by viewing the signals emitted by the Neurons. The maximum voltage obtained from the Combined output of the Neurons indicate the extent of damage to the structure caused by the Acoustic Emission. Depending on the level of voltage obtained the amount of Force induced into the structure can be calculated.
2 Simulated 2 by 2 Results F i r i n g o f t h e N e u r o n S w i t c h e s i n t h e X - D i r e c t i o n Neuron Number 1 3 0 0 0. 2 0. 4 0. 6 0. 8 1 1. 2 1. 4 1. 6 1. 8 2 T i m e ( S e c ) x 1 0-3 F i r i n g o f t h e N e u r o n s w i t c h e s i n t h e Y - d i r e c t i o n 4 0. 2 0. 4 0. 6 0. 8 1 1. 2 1. 4 1. 6 1. 8 2 T i m e ( S e c ) x 1 0-3
Simulation of 2 by 2 system Voltages from the Neurons 0.06 0.04 0.02 Voltage (Volts) 0-0.0 2-0.0 4-0.0 6 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 Tim e x 10-3
Experimental 2 by 2 Neural System Panel Fiberglass composite 2 Row Neurons 2 column Neurons Wave Speed was determined experimentally Wave Speed = 3000 m/sec
Experimental Set-Up 1. 8 PZT patches for sensing 2. 1 Pencil to simulate Acoustic Emission. 3. KROHN-HITE HITE Multichannel filter (MODEL-3905 B) 4. Lecroy 4 channel Oscilloscope (LT344 ) 6. 1 LAPTOP for Data Acquisition.
V1(Volts) 0.2 0 Experimental Results -0.2-4 -3-2 -1 0 1 2 3 4 5 0.2 x 10-4 V4 V3 V2 V2(Volts) 0-0.2-4 -3-2 -1 0 1 2 3 4 5 0.2 x 10-4 V4 V3 V2 V1 V3 (Volts) 0-0.2-4 -3-2 -1 0 1 2 3 4 5 0.2 x 10-4 V1 V4 (Volts) 0-0.2-4 -3-2 -1 0 1 2 3 4 5 Time(Sec) x 10-4 Location of Lead Break (36,18) inches
Wave arrival times at Neurons Case Neurons Experimental Sec Simulation Sec 1. V1 and V2 1.82 e -4 1.12 e -4 2. V3 and V4 3.04 e -4 2.96 e -4 1. The differences are due to change in sensor locations.
Equations for Detection of Crack (X-x 1 ) 2 + (Y-y 1 ) 2 = C *(t) 2 (X-x 2 ) 2 + (Y-y 2 ) 2 = C *(t+ t 2 ) 2 (X-x 3 ) 2 + (Y-y 3 ) 2 = C *(t+ t 3 ) 2 where (x 1,y 1 ) and (x 2,y 2 )and (x 3,y 3 ) = Sensor Locations C = Wave Speed X,Y = Simulated Acoustic Emission Location t = Time for the wave to arrive at the closest sensor
Case 1. Determination of Crack Location Actual Excitation Location (21,18) Predicted Excitation Location (22.74,21.99) 2. (12,18) (13.08,20.81) The maximum error is 22%. This error could be reduced 1. By increasing the number of sensors. 2. Exact wave speed might be different from the calculated wave speed ed 3. The sensors are modeled as one point, but in the experiment they are 1 inch apart. 4. Approximation in the wave arrival time and calculation of t in the experiment
Summary Both the experimental and Simulation results matched each other with a good amount of accuracy. Simulated Acoustic Emission was detected within a 4 inch radius on a 48 X 48 plate. This deviation in results could be reduced by decreasing the distance between two adjacent sensors. Can work well for large structures as less data acquisition channels are required.
5. CONCLUSIONS The ANS offers: Distributed sensing Parallel processing Continuous monitoring of the condition of the structure to prevent damage Detect Propagating Damage Warn of overstress and anticipate failure
Proposed Future Work To calculate the force that was applied to the structure. To test the Neural System by inducing ambient vibrations into the structure. To extend the Wave Simulation Code for various types of Sensors (PZT s, AFC s and PVDF)
ACKNOWLEDGMENT This work is supported by the U.S. Army Research Office under contract grant number G DAAD 19-00-1-0536; the NSF Center for Advanced Materials and Smart Structures at NCA&TSU; and the NASA Center for Aerospace Research at NCA&TSU. The wave propagation code used to perform the simulations was developed by Dr. William N. Martin and Dr. Anindya Ghoshal. This support is gratefully acknowledged.