H13 tool steel, known for its excellent thermal stability, wear resistance, and high toughness, is widely used in hot working molds. However, during the casting and processing of mold steel, internal defects can occur, which become a critical factor limiting the performance and service life of mold steel. These internal defects not only affect the mechanical properties of the material but can also lead to severe consequences such as fractures and failures during mold use. Therefore, effective detection of internal defects in H13 mold steel is an essential step to ensure mold quality and safety.
Traditional defect detection methods for mold steel can identify surface and near-surface defects to some extent but are often unsatisfactory when it comes to detecting hidden core defects inside the material. In recent years, ultrasonic testing (UT) technology has shown unique advantages in the internal defect detection of metallic materials due to its non-destructive nature, high sensitivity, and strong penetration ability. As UT technology evolves, detection methods and algorithms specifically optimized for materials like H13 mold steel have been developed. Through precise analysis and processing of ultrasonic signals, researchers can accurately identify tiny internal defects within the material, locate them, and perform quantitative evaluations. Moreover, with the rapid development of computer technology and artificial intelligence, automated processing and intelligent recognition of ultrasonic testing data have become possible, greatly improving detection efficiency and accuracy.
Through in-depth analysis and innovative application of ultrasonic testing technology, a detection scheme for internal defects in H13 mold steel has been proposed. This scheme combines advanced ultrasonic signal processing techniques and intelligent recognition algorithms aiming to achieve rapid and accurate detection of internal defects in H13 mold steel.
1 Ultrasonic Testing-Based Detection Method for Internal Defects in H13 Mold Steel
1.1 Acquisition of Ultrasonic Energy Characteristics of H13 Mold Steel Core
To detect internal defects in H13 mold steel, ultrasonic probes are used to inspect the mold and obtain ultrasonic energy characteristics. To ensure adequate penetration and resolution, probes with a frequency above 3.5 MHz are selected as the main detection equipment. To ensure tight contact between the probe and the workpiece surface and minimize air gaps that reduce ultrasound reflection and scattering, a specialized coupling agent is applied on the surface of the inspection area to enhance acoustic coupling between the probe and the workpiece.
During propagation, ultrasonic waves may be interfered with by various noises from inside or outside the material, reducing the signal-to-noise ratio and making defect signals difficult to identify accurately. The empirical mode decomposition soft-threshold method is used for ultrasonic noise reduction of the internal defect part of the H13 mold. Then, kernel principal component analysis is utilized to extract the ultrasonic energy characteristics of the internal defects of H13 mold steel based on the acquired ultrasonic energy features.
1.2 Analysis of Time-Domain Feature Information
For the extracted ultrasonic energy characteristics of the core of H13 mold steel, time-domain feature information is extracted using a time window function. The time window function smooths the signal in signal processing, reduces spectral leakage, thereby improving the accuracy of signal analysis. The formula is as follows:
Where Ta represents the extraction amplitude of the time window function, t0 represents the center of the time window, ωt represents the width of the time window, and t1 represents the displacement vector of the time window. Based on the time window function described above, the time-domain feature information analysis expression of the ultrasonic signal is constructed as follows:
Where ε represents the offset of the time window function, x* represents the conjugate information signal associated with the echo information, Z(t,ε) represents the instantaneous correlation function of the echo information data, and ω represents the signal frequency.
During the process of parsing the time-domain feature information, there exist interference vectors caused mainly by signal noise and distortion. When signals are transmitted and distorted or interfered by different signal sources, signal overlap occurs, producing distributed interference terms. To eliminate these interference terms, an instantaneous correlation function is introduced into equation (2) to suppress the interference degree of signal superposition. The instantaneous correlation function of echo information data is given by:
Where Za(t,ε) and Zs(t,ε) represent the autocorrelation vector and the conjugate correlation vector of the echo information in the ultrasonic signal, respectively. Equation (3) allows the removal of excessive distributed interference items from the extracted time-domain feature information, ensuring the accuracy of the extraction results.
1.3 Detection of Internal Defects in H13 Mold Steel
Based on the extracted time-domain feature information, support vector machines (SVM) are employed to convert the defect detection problem into an optimal classification plane solution problem to achieve defect classification detection. Thus, the ultrasonic time-domain feature information serves as input data x(t) for the SVM, and the sample set of ultrasonic characteristic information for the core of H13 mold steel is constructed as (vj,uj), where vj and uj represent the classification attributes and labeling information of the time-domain feature information, respectively. The principle of using SVM for core defect detection lies in projecting the input sample set into a high-dimensional feature space and setting up the optimal hyperplane for defect detection, resulting in classification detection outcomes. An optimal classification hyperplane is thus constructed, with the algorithm given by:
Where m is the number of samples of time-domain feature information, cj is the Lagrange multiplier, k(v,vj) is the kernel function of the SVM, a is the penalty compensation factor, and w is the weight bias compensation term. This hyperplane maximizes the boundary between different categories while minimizing differences within the same category. By constructing an initial feasible domain and setting the training objective of the optimal hyperplane to classification error, the defect detection model is formulated as:
Where oj represents the classification detection result of the defect, δMES is the training error of the SVM, μj is the slack variable, and o’j is the actual internal state information of the mold extracted by ultrasonic testing technology. Through the aforementioned steps, classification detection of internal defects in H13 mold steel can be completed. By combining SVM to construct an optimal classification hyperplane and setting the plane’s training objective to classification error, a defect detection model is built. Combining this section with the previously mentioned echo information acquisition and time-domain feature information analysis completes the design of the ultrasonic testing-based detection method for internal defects in H13 mold steel.
2 Experimental Verification
To prove that the proposed detection method outperforms conventional internal defect detection methods for H13 mold steel in practical defect detection, after completing the theoretical design, an experimental phase was constructed to examine the detection effectiveness.
2.1 Experiment Description
To verify the superiority of the proposed method in practical defect detection, two groups of conventional internal defect detection methods for H13 mold steel were selected as comparison objects: one based on machine vision and another based on perception encoders. An experimental platform was built to apply three detection methods to the same H13 mold for defect detection and compare the actual detection effects of different methods.
2.2 Experimental Subjects
To obtain a dataset of internal defect samples in H13 mold steel, scans were conducted on molds of different specifications to collect echo signals of various defects. Preprocessing was performed on the collected signal data to ensure the quality and reliability of the original dataset. Amplification and gain adjustments were appropriately made to improve the signal amplitude so that weak defect signals could be better detected. Extracted features were formatted to adapt to subsequent machine learning algorithms and models.
The constructed defect sample dataset includes multiple samples, each containing several features. Features are selected based on the characteristics of ultrasonic signals and defect types, such as amplitude, frequency, phase, arrival time, etc. Each sample also contains a label indicating whether the sample contains a defect and the type of defect. The dataset is stored in a structured format for ease of subsequent data processing and analysis. Each sample is represented as a vector containing multiple feature values and labels. Labels are annotated in a classified or labeled manner to distinguish different types of defects.
In the sample defect data, to accurately compare the actual detection effects of different methods, the positions of defects were simulated in advance for this experiment, yielding localization results for different defects. This positioning information serves as the detection standard to compare the accuracy of different methods, as partially seen in Table 1.
By applying three methods to simulate the detection of the experimental dataset, once the detection is complete, the defect classification detection results of different methods are recorded and compared with the actually simulated positioning information to effectively compare the detection performance of the methods.
2.3 Comparison Results of Defect Diagnosis Precision
The experiment uses the recall rate of defect detection results under different methods as a comparative index to measure the actual detection accuracy of different methods. After experimentation, it was found that the highest recall rate for conventional detection method A was 0.47%; for conventional detection method B, it was 0.63%; and for the adopted detection method, it was 0.73%. It can be seen that the detection accuracy of different methods varies when simulating detection on the same experimental dataset. The recall rates corresponding to the two conventional detection methods are significantly lower, with the highest recall rate being only 0.65%, whereas the ultrasonic testing-based detection method for internal defects in H13 mold steel possesses higher detection accuracy, with a maximum recall rate of 0.73%, proving that this method offers more ideal detection results.
3 Conclusion
A new ultrasonic testing-based detection method for internal defects in H13 mold steel shows certain effectiveness in detecting internal defects in H13 mold steel. However, there are still some shortcomings in practical applications that need further improvement and perfection. Future research should delve deeper into noise reduction processing of ultrasonic signals.
