·Home ·Table of Contents ·Reliability and Validation 2

The Artificial Intelligence in Service of Ultrasonic Inspection Reliability F. Bettayeb Scientific Research Center on Welding And Non Destructive Testing, C.S.C, Route de Dely Brahim, BP: 64, Chèraga, Algiers. Tel/fax: (213-2) 361850. Email:bettayeb@excite.com,_bettayeb@email.com H. Benbartaoui , B.Rouaroua University of Science and Technology USTHB, El Alia BP:32, Algiers. Contact

Abstract

Increasing automation of NDT inspection procedures has conducted to an increasing quantity of acquired data, which must be interpreted in a short amount of time. Therefore, for cost-effectiveness, the process necessitates reliable automated data interpretation techniques.
In this work we develop an intelligent tool, which proceed with an automatic classification algorithm, and a pattern recognition scheme of A-scan data. After a defects classification based on shape/amplitude, the system processes through learning, thanks to an automatic apprenticeship stage.
A previous study on learning systems techniques, like the expert systems, case-based reasoning systems, fuzzy sets and fuzzy logic, hybrid systems etc., has concluded that: for NDT data interpretation, the main advantage of case-based reasoning systems [1], is their ability to learn rapidly from classification made by the operator, who will transfers automatically his reasoning with the help of the cognition engineer. Among these systems, Artificial neural networks (Ann) are established to be the most suitable classifiers for pattern recognition problems. Since most NDT inspection procedures need the presence of an operator, this method allows the proximity of the operator without disturbing the system. So that, the obtained system will be able to perform classification, of the different events originating from defects, noise and component geometry.
Key words:Ultrasonic testing, A-scan signal, classification, feature selection, learning, neural network.

1. Introduction

The application of Ndt techniques to estimate material quality depends on a large body of knowledge based on operator's competence and experience. In manual testing particularly, the identification of relevant from non-relevant indication as well as defect characterisation, is highly examiner dependant [2]. The volumetric/non-volumetric type of indication detected during an ultrasonic weld inspection has a decisive influence on weld acceptance or repair.

Usually, during the examination procedure the ultrasonic operator is alone, depending only on his own observation of the signal features, such us echo shape, amplitude level, defect position within the joint geometry, rotation of the transducer around the defect location, etc. These parameters to which, for some of them no figure can be put, are combined by the operator in a non explicit way to lead to the diagnosis. In the aim of automation, this interpretation practice found a natural extension in the exploitation of artificial intelligence computer tools.
Usually automated data interpretation will be done using statistical or artificial intelligent techniques. Among these tools, the artificial neural networks have been verified to be very efficient in the area of pattern recognition, since the current study formalism is dealing with a signal shape recognition combined with heuristics rules. The system drawn here by the artificial neural networks method, is based on an automatic classification and learning algorithm of A-scan data from defects, thanks to an automatic apprenticeship stage.
For accuracy and effectiveness needs, our classification scheme has been accompanied by an experiment ultrasonic inspector with a qualification of a level 2 certified by the IAEA agency.

2. Neural network background

Artificial Neural networks (Ann's) have been used widely in many application areas in recent years. An artificial neural network (Ann) consists of a network of nodes connected via adjustable weights. The weights can be adjusted so that a network learns a mapping represented by a set of example input/output pairs. An Ann can in theory reproduce any continuous function Fⁿ (r)F^m, where n and m are numbers of input and output nodes [3].
Most application use feed forward Ann's and the back propagation (BP) training algorithms. There are numerous variants of the classical BP algorithms which assume a fixed Ann architecture [4]. They train only weights in the architecture that includes both connectivity and node transfer function. There have, many attempts in designing Ann architecture automatically, such as various constructive and compacting algorithms [5].
Generally a constructive algorithm starts with a minimal network, and adds new layer, nodes and connections if necessary during training, while a compacting algorithm does the opposite, i.e. deletes unnecessary layers, nodes and connections during training [6].

Neural networks are nowadays predominantly applied in classification tasks. The neural network impose strong requirements on the data and the inspection, however when these are accomplished then good automatic classification systems can be developed. The architecture of a back propagation network is used due to the fact that it is the most robust and common network. Networks used for classification have commonly as much input neurones as there are features and as much output neurones as there are classes to be separated [5] [7]. In the back propagation algorithm which is based on a gradient descent-method, each neurone of a layer is connected to each neurone in the previous and in the next layer.

3. Problem statement

Drawing an automatic recognition of weld defects requires a powerful formalism. the chosen methodology is based on the fact that the classification is considered as a closed system, with well known inputs and outputs. A general class of such a system can be represented by the following state space model:

(1)

In the case of ultrasonic system the inputs are represented by the shape envelope of the A-scan defects and the outputs are 2 classes about planar and volumetric defects.

3.1. Ultrasonic data features for defect recognition.
A detailed knowledge of the interaction between ultrasonic waves and defects, the propagation medium and the conditions in which ultrasonic investigations are curried out, are all elements of basic importance in defect recognition. The strategy is to extract some parameters, enabling the featuring of the pulse echo envelope reflected from a defect, on A-scan images obtained using a single transducer, relative both to the maximum reflection transducer position and to the transducer-defect distance variation. Some of the recognition rules for the most important defects are drawn in the table below.

Defect Type	Echo form	Echo form after transducer movement from the defect location
		Mvt // to the joint	Mvt ^ to the joint	Around the joint
Porosity	With jagged aspect	The echo disappears rapidly	The echo declines quickly	The echo is nearly constant.
PorosityGroup	Set of small echoes with stiff fronts	The global echo persists, with mobility of the sub echoes position.	Same conditions as // movement.	Same conditions as the other movements.
Slag Inclusion	has an irregular front	The echo persists but its shape changes	The echo go away rapidly	The echo decreases gradually
ImperfectPenetration	The echo is high with stiff front.	The echo behaviour subsists	As for // movement	The echo disappears
Lack of fusion	As for lack of penetration	Same conditions.The only	Same conditions differentiation is	Same conditions. about the defect position.
Cracks	High echoes superposition, with stiff front	The echo persists	The echo persists	The echo reduces and raises without going out
Table 1: "Defect recognition rules from manual testing"

4. System architecture :

Neural network classifier
Choosing the architecture of a neural network for a particular problem usually requires some prior knowledge of the problem's complexity. The network topology, however directly affects the 2 most important factors of neural network training. Both theoretical studies and simulations, show that larger than necessary, networks tend to overfill the training data and thus have poor generalisation, while too small a network will have difficulty learning the training samples [8]. Currently, there are no formal methods to customise or select the right network structure. Here the Ann classifier is trained to represent some decision between the 2 classes to be recognised. The system architecture is drawn in the following figure.

Fig 1: "Network architecture"

The input layer consists of the envelop shape of the defect signal called the defect map acquired from a numerical oscilloscope on 1024 samples in a vectorial representation. And the output layer is a Boolean representation of planar or volumetric defect decision represented with 2 classes.

5. System strategy

After initialising the neurones of the network, the learning starts with applying the training vector to the network input. This vector is processed through the different layers. At the network's output the error between the actual and the desired output is measured. This error is afterwards minimised by back propagating it through the network and reconfiguring the neurones and their weight's matrices in the whole network. This is repeated for all training patterns until a special termination or convergence criterion is met, in our case a small error.

5.1. Learning process
The algorithm starts with random initial weights and learns with different maps, respectively. All of the connections in the network are adaptive and are trained with back propagation algorithm. Since the back propagation algorithm requires differentiability along the signal path of the network, we adopt as a transfer function at each unit the familiar 'Sigmoid' function as follows:

(2)

Where: a : is the weight; x : is the input of the unit.
b : is randomly selected in [-0.5 , 0.5] at start state and modified during training.

So, when x is presented to the network and propagated forward, the state x_j in each layer is:

(3)

Where: b_j : is the weight of x_j , connecting x_i of previous layer to x_j of the current layer.
And nl : the number of nodes per layer.

After learning each layer will have different internal representation.

5.2.Objective function
In order to evaluate the weights, they have to be trained from a random set of initial weights, using the training back propagation algorithm, with a minimising process of an objective function. This has be done by the choice of the conventional mean squared error function (MSE), between the actual output x_j and the desired one x_d.

Minimising an objective function is usually performed by a gradient descent procedure, which requires the computation of the gradient of the objective function with respect to connection weights, and back propagation is just an efficient way to compute this gradient [5].The modification of weights is performed as the ensuing algorithm.

Algorithm:


• The objective function is:

  (3)

• To minimise E (x), each weight is updated by:

  (4)

  Where : D : is the gradient; n : is the training index; b: weighting connection; 0 £e£1
  And a: error momentum e [0.05 , 0.25]
• So for each layer 'l':

6. Conclusion

Since Ndt is an "expert" oriented field, the ability of learning systems to offer a framework for emphasising the manner in which decisions are made, and then to formalise the path by which they are reached is most important for automation purpose. Feature extraction is a very important aspect of solving the problem of pattern recognition, and Neural networks of various types and structures have been found to be efficient tools for identifying such systems.

In this paper we have worked on artificial flaws from steel plates, and on some natural defects from welds, for which the obtained signals are sampled on 1024 points. And for each defect location, we have taken several signals dealing with different transducer positions. The learning is performed on about 90 patterns. Future work concerns the search of an optimal configuration of the global network in which a sub net of each defect family will be included, so that to obtain a more categorical apprenticeship stage.

Acknowledgements

The assistance of Mr M. Hassam inspector in Ndt, in ultrasonic defect detection and signal acquisition and his contribution in the building of the recognition rules are greatly appreciated. Also we are thankful to Mr A. Haddad for his signal acquisition software.

References:

1. P.P.Van't Veeen, J.Jarmulak, "Automated interpretation of heterogeneous Ndt measurements", in the 7^th ECNDT proceedings, Vol.3 , pp. 2355-2362, 1998.
2. A.Gayer & al., "Automatic recognition of welding defects in real time radiography", NDT international, Vol23 , N°3, Jun 90.
3. G.E.G Benoggi & al., " The effect of reasoning logic on real time decision making", IEEE transactions on systems, man and cybernetics, vol. 27, N°6, Nov 97.
4. T.P. Minka & al.," Interactive learning with a 'society of models' ", pattern recognition , Vol. 30 N°4, April 97.
5. X. Yao, Y.Liu, " A new evolutionary system for evolving artificial neural networks", in IEEE trans Neural networks, May 97, Vol. 8, N°3 pp: 694-713
6. R.Reed, " Pruning algorithms - a survey", in IEEE trans. Neural networks, Vol.4, 1993, pp. 740-747
7. I.K. Sethi, "Neural implementation of tree classifiers", in IEEE trans. systems, man and cybernetics, Vol. 25, N°8, August 95, pp. 1243-1248.
8. W.L.AN. Whitlow, L. Jeffrey & al.," Neural network modelling for a dolphin's sonar discrimination capabilities", in the journal of acoustical society of America, Vol.98,N°1, July 95,pp.43-49.

© AIPnD , created by NDT.net

|Home|    |Top|