R. Sikora, T. Chady, P. K. Frankowski The purpose of this paper is to present knowledge extraction algorithms, dedicated for new electromagnetic system used to evaluate steel bars in reinforced concrete structures. All stages of the rebar identification process have been presented. At the first step, relations between parameters of the tested structure and measured waveform are extracted. For this purpose, a dedicated association rules learning algorithm is proposed. In the next stage, the collected data are filtered and smoothed. Finally, classification models are implemented, tested and evaluated. The experimental verification of the applied techniques was carried out, and the selected results are presented. Department of Electrical and Computer Engineering West Pomeranian University of Technology, Szczecin, Poland d 60 d 30 d x U 30 U 60 U 0 U min U max X 60 X 30 X max X min X 0 START k=1, , Scan the database part Y to get the pair of instances varying only one parameter. Compare all attributes (part X) and save in the database D the relations between items as one of the three class (Ŷp ј, љ,-). D h Reinforcement bar Concrete Sinus Function Synthesizer I Power Amplifier I Lock-In Amplifier Sinus Function Synthesizer II A / D Converter Control Computer Power Amplifier II Data Understanding Problem Understanding & Preparation Data Preparation Modeling Evaluation Deployment BLOCK DIAGRAM OF THE MEASURING SYSTEM EXAMPLE OF THE MEASURED WAVEFORM AND DEFFINITION OF THE ATTRIBUTES. ABSTRACT: Two kinds of attributes are used in the analysis: shape factors d and maximal value of the waveform U max . Value of the shape factors d are independent of the waveform maximal value. Definitions of these factors are explained in a Figure (U 30 =30%∙U max ,U 60 =60%∙U max etc.). In the identification process 96 attributes were used. Transducer position Voltage amplitude Example of the signal measured during movement of the transducer over the concrete structure with the single rebar KNOWLEDGE EXTRACTION PROCESS Presented process is based on a CRISP-DM model Y X D h class U max d x d 98 … 10 10 1 12 3 5  10 15 1 10 4 7         14 50 3 2 7 9  Y X D h class U max d x d 98 … - ј - љ ј ј         - - ј љ - -  ј - - љ ј ј  Generate k-itemset candidats (B). k=1 => If(h ↑) Then (U max ↓),… k=2 => If(h ↑) Then (U max ↓& d x ↑),… etc Check support.   ) ( ) ( # : # supp A P A D i d A D i d     For each candidate: supp≥suppŵiŶ ? supp.min is given by user The set of k+1 candidats is ? Check confiedence. For each itemset F: coŶf≥coŶfŵiŶ ? conf.min is given by user Add to frequent itemset F B=F   ) | ( ) ( supp ) ( supp conf A B P A B A B A     Add to association rules and STOP ASSOCIATION RULE LEARNING ALGORITHM Yes No Yes Yes If(h ↑) Then (U max ↓& d x ↑) [supp=11%, conf=100%] etc. Presented method is based on an Apriori algorithm Y A  X B  Knowledge Extraction Algorithms Dedicated for Identification of Steel Bars in Reinforced Concrete Structures k=k+1