Monolithic zirconia (MZ) crowns have become widely popular as the material of choice for dental crowns, especially for restoring severely damaged teeth [
1,
2,
3]. High-quality aesthetics, biocompatibility, and mechanical strength are recognized attributes of MZ crowns. As a result, MZ has been gradually replacing conventional ceramic systems as the favored restorative material. A critical factor in the processing of MZ is the development of computer-aided design and computer-aided manufacturing (CAD/CAM). Dental crowns for root canal-treated teeth and implant prostheses, as well as dental bridges, can all be fabricated using CAD/CAM technology, which has gained widespread use. As such, to reduce the chance of failure, it is essential to identify and localize cracks in these crowns as soon as possible. Furthermore, mechanical loading and structural problems can lead to crown failures. MZ crowns are fabricated using a multi-step, intricate process that can result in a variety of surface flaws [
4,
5], particularly when subjected to extreme stresses such as those brought on by bruxism [
6,
7,
8]. Currently, several nondestructive methods are available for detecting flaws, including direct visualization [
9], fiber-optic transillumination, combined tactile examination [
9], radiographic examination, computer tomography [
10,
11], and acoustic emission (AE) testing [
12]. In clinical evaluations, dental crown removal is not feasible, whereas destructive testing methods, such as Vickers indentation and fractography, are limited to laboratory settings [
13,
14]. The subtle nature of cracks in dental crowns render such defects challenging to detect, leading to difficulties in diagnosis by clinicians. Misdiagnosis can result from visual fatigue or judgment errors. Flaws may be missed if the X-ray beam does not align with the defect, as seen with cone-beam computed tomography (CBCT), which is prone to streaking artifacts from radiopaque materials [
11]. With voxel sizes of 75–400 μm, CBCT often fails to detect finer microcracks [
15]. Noise and interference are nearly unavoidable, and limited resolution may obscure microcracks, especially in early detection. Micro-computed tomography (micro-CT) often suffers from artifacts that degrade image quality, and its high resolution requires radiation doses in SkyScan 1272 workflow of up to 141.4 kVp, far exceeding the clinical CBCT 60–90 kVp [
16]. To combat such disadvantages, AE analysis emerges as a potential alternative method. AE testing is a widely used passive nondestructive testing (NDT) method that evaluates stress waves generated by cracking within a material [
17]. It detects the spontaneous release of strain energy from microscopic or macroscopic changes, converting it into an electrical signal for further analysis [
18]. These emissions can be released from localized stress concentrations, sudden material changes, crack propagation, or other dynamic processes. As a nondestructive technique, AE testing measures and examines the elastic energy released during crack formation. AE testing is frequently employed in structural health monitoring (SHM) to assess the integrity of storage tanks [
19,
20], railways [
21], aircraft parts [
22,
23,
24,
25], welding processes [
26], medical devices such orthopedic joint prostheses [
27,
28], and dental materials [
29,
30,
31,
32,
33].
SHM therefore offers a more dependable and affordable method for the maintenance of dental restorations, as it enables early detection and prevention of cracks that may impact the structural integrity of dental crowns. Accurate fracture localization and analysis are crucial for improving the routine practical inspections of dental restorations. Noninvasive techniques in AE technologies, known as modalities, are preferred due to their nondestructive nature. AE is particularly effective at isolating individual components [
19,
34]. The ability of AE testing to monitor real-time debonding kinetics at the tooth-composite/prosthesis interface offers valuable insights into the effects of interfacial debonding [
35,
36,
37,
38,
39]. AE testing can help identify damage progression during fatigue tests by analyzing location, energy, frequency, and waveform direction [
40]. Therefore, we propose that AE testing can prove to be a superior technology for dental material analysis. High-resolution spatial signals with denoised information can be used to develop automated diagnostic tools for dental practitioners to identify and analyze cracks with higher efficiency and accuracy. Early diagnosis and classification of faults in restorations are essential to improve treatment outcomes and increase the survival rate of dental restorations. At the same time, artificial intelligence-based methods can provide quick decision-making without human intervention, and deep learning (DL) holds great potential for the evaluation of zirconia combined with nondestructive testing–acoustic emission (NDT-AE) [
18]. Crack recognition in dentistry can be challenging and tends to rely heavily on automated diagnostic systems. This study focuses on identifying cracks on crown surfaces during clinical examination, with the aim of developing sophisticated techniques for the classification of cracks. A study by Lee, D.W. et al. showed that the automated CNN architecture using radiographic image analysis techniques had both limitations and future potential. All three DL algorithms achieved an AUC of over 0.90 in detecting fractured dental implants, with the automated Deep Convolutional Neural Network using periapical images outperforming both the modified VGGNet-19 and GoogLeNet Inception-v3 [
41]. The advancement of technology has led to the development of acoustic emission (AE) detection for restorations, which can identify the position of cracks. Automated or computerized methods are expected to reduce manual diagnostic errors and provide precise data for accurate diagnosis and treatment planning. Current crack detection methods transform AE signals to continuous wavelet transform (CWT) for more accurate result interpretation. AE-NDT employing CWT images allows for multi-class classification to precisely identify crack locations, particularly those close to adjacent teeth in the oral cavity. This study investigated crack detection at five vulnerable areas of the crown: incisal (cutting edge), left, right, labial (anterior), and palatal (posterior) surfaces.
This study utilized the Enhanced Inception-ResNet-v2 model, combining the advantages of the Inception and ResNet models. This model employs multi-scale depth-wise separable convolution for efficient and fast feature extraction, and includes a feature extractor to preserve important formation, accelerating convergence and filtering out less relevant data. The model eliminates redundant parameters to maintain controllable complexity and reduce computational load. Compared to existing algorithms, Inception-ResNet-v2 demonstrates superior accuracy and stability [
42]. For image classification tasks, the Inception-ResNet-v2 model that is suggested has been shown to be quite effective. The model handles multiple scales of information from various input images, aiding in diagnosing diseases. By leveraging the combined strengths of Inception and ResNet architectures, the model enhances training scope, robustness, and depth, achieving superior efficiency and stability. The Inception-ResNet-v2 model demonstrates higher classification accuracy than other existing methods, making it an effective tool in machine learning for image classification problems [
43]. This research aimed to enhance crack detection and localization in MZ crowns utilizing CWT to generate 2D images from AE signals. These images were analyzed using CNNs to analyze the AE-CWT patterns, specifically the Inception-ResNet-v2 architecture. CWT preprocessing was applied to raw AE signals. Images were processed with AE technology produced by pencil lead breaking (PLB) to classify crack sites in MZ crowns into one of the five specified categories. CNNs automatically extracted features through various convolutional layers, addressing the challenges of crack classification without manual split regions and feature extraction. Consequently, the extracted features were expected to yield robust results for subsequent classification. Utilizing the Inception-ResNet-v2 architecture, the AE-CWT patterns were analyzed to localize and classify cracks into one of five specified categories. This research aimed to contribute to the dental field by providing a classification system for cracks in zirconia crowns, thereby facilitating their identification and management in clinical practice.
This research is organized as follows: the introduction and review of several earlier studies that used the AE and DL models for the classification of dental material detection are included in
Section 1. The data and methodology utilized in this work are described in
Section 2, with a particular emphasis on the deep learning architecture. The Inception-ResNet-v2 analysis results are shown in
Section 3, accompanied by an assessment of the effectiveness of the crack classification system.
Section 4 presents the discussion. The conclusion of this study and recommendations for future research are provided in
Section 5.