Non-Destructive Assessment of Internal Wood Conditions Using a Developed Electrical Capacitance Tomography–Dielectric Spectroscopy System

Structural abnormalities in trees often manifest as internal defects such as decay and cavities, which are typically caused by biotic agents, including wood-degrading fungi, bacteria, and wood-boring insects such as beetles and termites [1]. Fungal pathogens, particularly those capable of degrading lignin and cellulose, initiate decay by infiltrating the wood and breaking down its primary structural components. This leads to progressive softening of the tissue, discoloration, and a notable decline in mechanical integrity [2-4]. In parallel, termites and similar insect pests contribute to internal cavity formation by excavating extensive tunnels through the heartwood, creating hollow zones within the tree [5]. These internal deteriorations, whether due to microbial decay or insect activity, substantially impair the tree's structural strength and increase its susceptibility to failure, particularly under mechanical or environmental stress.

The evaluation of internal wood conditions is essential for both forest management and the timber industry, as internal defects such as decay, cavities, and structural degradation can significantly reduce wood quality and mechanical performance. Early detection of these defects is therefore crucial to ensure safe utilization and efficient resource management. However, conventional inspection methods are often destructive, time-consuming, or rely heavily on visual inspection of external symptoms, which may not accurately represent the internal condition of wood logs [6, 7].

To address these limitations, various non-destructive evaluation (NDE) techniques have been developed to assess internal wood properties. These include acoustic or stress-wave tomography, X-ray computed tomography, micro-drilling resistance testing, and ground-penetrating radar [8-14]. While these techniques have demonstrated effectiveness in detecting internal defects, some require expensive equipment, complex operational procedures, or limited portability for rapid field inspection.

Electrical and dielectric-based sensing methods have recently emerged as promising alternatives for wood condition monitoring. The dielectric properties of wood are strongly influenced by factors such as moisture content, density, and internal degradation processes. As a result, dielectric measurements can provide useful information for identifying internal defects and structural heterogeneity. Dielectric spectroscopy, which analyzes the frequency-dependent dielectric response of materials, has been widely used to characterize biological and lignocellulosic materials [15, 16].

Another promising technique is Electrical Capacitance Tomography (ECT), which reconstructs the spatial distribution of permittivity inside an object using capacitance measurements obtained from multiple electrodes surrounding the material without requiring direct contact with the specimen [17-20]. Although ECT has been widely applied in industrial process monitoring, its application in wood diagnostics remains limited. This limitation arises partly from the complex electrical behavior of wood, which is highly anisotropic and strongly influenced by moisture content and internal structural heterogeneity [21, 22]. These factors can cause significant variations in dielectric properties, making the interpretation of capacitance measurements more challenging when using conventional single-frequency ECT approaches.

To address these challenges, integrating ECT with dielectric spectroscopy offers a promising alternative [23]. Spectral measurements across a range of frequencies allow the characterization of the complex permittivity of wood, expressed as:

$\varepsilon^*=\varepsilon^{\prime}-j \varepsilon^{\prime \prime}$        (1)

where, ε' represents the dielectric constant and ε'' corresponds to dielectric losses [24]. Variations in moisture content, density, and internal degradation processes affect these dielectric parameters differently across frequencies [16]. Therefore, combining tomographic sensing with spectral analysis can provide richer diagnostic information, enabling improved differentiation of internal wood conditions such as intact, decayed, or hollow regions [25].

Based on this concept, this study proposes the development of an integrated Electrical Capacitance Tomography and Spectroscopy (ECT-S) system for evaluating the internal condition of wood logs. The system measures dielectric responses across a wide frequency range (250 kHz–60 MHz), and the resulting impedance spectra are analyzed using Principal Component Analysis (PCA) and machine learning classifiers to distinguish between intact, decayed, and hollow logs.

3.1 Impedance patterns and resonance behavior of wood logs

The capacitance-based measurement system produced raw complex data from the wood logs, primarily composed of imaginary components. As a result, data extraction procedures were necessary to calculate impedance, resistance, and capacitance values across the frequency range of 250 kHz–60 MHz. Subsequent analysis centered on identifying resonance behavior, which consistently manifested as a pronounced increase in impedance within the 17–30 MHz range (Figure 5), while lower impedance values were observed outside this band. Notably, the raw dataset included both outliers and missing values, likely introduced by measurement inaccuracies related to environmental influences, sensor limitations, or system noise. Therefore, data preprocessing, comprising outlier detection and imputation of missing values, was essential to ensure the integrity of the measurements and to accurately represent the material’s dielectric polarization in response to the applied electric field [31].

Figure 5. Impedance graphic of wood logs compared with air

The analysis focused on the resonance frequency range, where peak impedance values were used to assess the wood’s dielectric response. The results revealed clear distinctions between different wood conditions. Intact segments consistently showed the highest impedance peaks, indicating strong dielectric polarization, while hollow sections produced values that were very close to those of air, suggesting minimal interaction with the electric field. In this context, air was used as a baseline reference because of its low relative permittivity, approximately 1, in contrast to dry wood (which typically ranges from 2 to 4) and wet wood, which can exceed 20 due to its higher water content [24, 32, 33]. These findings highlight how the dielectric response can reflect underlying structural integrity, making it a useful indicator for assessing internal wood conditions.

The impedance spectra of the six wood logs, along with air as a reference standard, revealed distinct clustering patterns (Figure 5). Log 6 fell within a high-impedance cluster, characteristic of intact wood exhibiting strong resonance behavior. In contrast, Logs 4 and 5 formed a mid-impedance cluster, with downward-shifted peaks, suggesting the presence of mild to moderate decay or small internal voids. Meanwhile, Logs 1, 2, and 3, along with the air reference, were grouped into a low-impedance cluster, marked by flat spectral profiles. These curves closely approached the air baseline, consistent with the behavior expected from hollow or severely deteriorated wood, or loosely packed material with minimal dielectric response.

Although all specimens exhibited resonance within the 17–30 MHz range, the location of the resonance peaks varied. Shifts to higher frequencies were generally associated with denser or drier wood, while shifts to lower frequencies appeared to correlate with higher moisture content or structural degradation. Notably, the similar spectral profiles observed in the decay logs (Logs 4 and 5) suggest that a single impedance spectrum may not be sufficient to distinguish between different stages or types of decay. This highlights the potential benefit of applying multivariate analysis techniques, such as PCA or clustering algorithms, to better classify and interpret wood decay characteristics across a spectrum of conditions. The air spectrum serves effectively as a control, representing the lower bound of permittivity and impedance response, and is clearly separated from all wood samples.

Figure 6 presents the impedance magnitude spectra for each individual log segment, highlighting the frequency-dependent electrical behavior across varying internal conditions. The intact logs (e.g., Log 4, Log 6) consistently show high and stable impedance in the 17–30 MHz range, with gradual slopes and minimal fluctuations. These smooth profiles are characteristic of dielectric continuity, reflecting structurally sound wood with evenly distributed moisture and intact cellular architecture. The impedance peaks correspond to high capacitive opposition to current flow, as expected in solid and homogenous dielectric media.

Figure 6. Impedance graphic of each log and segment compared with air as reference (a) Log 1, (b) Log 2, (c) Log 3, (d) Log 4, (e) Log 5, and (f) Log 6

In contrast, hollow logs (e.g., Log 1, Log 3) display significantly lower impedance magnitudes across the entire spectrum. This reduction is especially evident in the high-frequency region (above 25 MHz), where impedance curves flatten and show a complete absence of resonance features. This behavior indicates the dominance of air-filled voids, which reduce the effective relative permittivity and disrupt internal capacitive networks. The lack of clear spectral transitions suggests limited field interaction with solid wood material, in line with severe structural loss.

Decay-affected logs (e.g., Log 2, Log 5) demonstrate intermediate impedance behavior, with irregular curve shapes and localized dips in the 20–30 MHz band. These features may represent partial resonance effects resulting from decayed tissues acting as lossy, inhomogeneous dielectrics. The presence of fluctuating impedance gradients implies heterogeneous moisture content and breakdown of wall-bound water retention in partially degraded wood regions. In several cases, a mild resonance signature appears near 23–27 MHz, possibly caused by cavity-like geometries formed during mid-stage degradation.

Taken together, the spectral profiles in Figure 6 provide a diagnostic basis for distinguishing between wood conditions through their dielectric signatures. The resonance behavior observed in decayed samples, the damping of impedance in hollow logs, and the preserved permittivity in intact wood support the feasibility of frequency-based feature extraction. These distinctions justify the application of PCA and supervised classifiers in subsequent sections.

Interestingly, a deviation from the theoretical relationship between electrical impedance and permittivity was observed. According to the fundamental equation below [24]:

$Z=\frac{1}{j \omega \varepsilon_0 \varepsilon_r A / d}$        (4)

where, Z is the electrical impedance, j is the imaginary unit, ω is the angular frequency of the AC signal, ε₀ is the permittivity of free space, ε_r is the relative permittivity, A is the area of the electrodes facing each other, and d is the distance between the electrodes.

It is commonly expected that impedance decreases as permittivity increases. Following this logic, one might assume that air-filled cavities, with their inherently low permittivity, would exhibit higher impedance values. While this reasoning is sound in a static context, experimental observations showed that hollow regions frequently produced lower impedance values. This apparent contradiction is explained by the influence of high-frequency resonance effects and dielectric loss mechanisms. As noted by Torgovnikov [24], dielectric resonance phenomena become particularly prominent at frequencies above 10 MHz and can significantly alter impedance behavior. Therefore, the sharp reductions in impedance observed at specific frequencies are not anomalies, but anticipated outcomes of these effects.

Additionally, dielectric loss plays a critical role in the energy dissipation process. It represents the portion of the electric field energy lost as heat, particularly in materials with polar molecules. Segments of wood that are dense, moisture-rich, or structurally intact typically exhibit higher dielectric loss due to dipolar relaxation involving both water molecules and the wood’s structural polymers (e.g., cellulose and lignin). This contributes to increased permittivity and correspondingly higher impedance. In contrast, air-filled cavities lack the molecular structure necessary to absorb and dissipate energy effectively, particularly at higher frequencies, resulting in a net decrease in impedance values [34-36].

These interpretations are further supported by the observed impedance behavior across the frequency spectrum. Notably, wood logs containing internal defects exhibited higher impedance than intact samples at lower frequencies (< 10 MHz). However, as the frequency increased, particularly within the 17–30 MHz range, a pronounced drop in impedance was observed. This behavior is consistent with known dielectric resonance and energy loss mechanisms, where structural degradation alters how the material interacts with the applied electric field. Such frequency-dependent patterns provide valuable diagnostic signatures, making them particularly suitable for spectroscopic classification and enabling differentiation of internal wood conditions with greater precision.

3.2 Image analysis results

Each log segment was evaluated using cross-sectional image analysis to identify internal defects, including regions of decay and hollowness. Defects were visually distinguished based on observable changes in texture, discoloration, and the presence of voids within the wood structure. These areas were then delineated as defective zones. The extent of deterioration was quantified using an areal ratio, defined as the area of the defect divided by the total cross-sectional area of the corresponding segment. Representative images of each segment, along with the visually identified defect zones, are provided in Table 1. Meanwhile, the quantitative results of the abnormal area ratio for each log segment, determined through ImageJ analysis, are summarized in Table 2.

Table 1. Picture of cross-section from each segment and log

Table 2. Ratio of abnormal area in wood cross‑sections per log and segment from ImageJ analysis

This investigation confirms the capability of the developed capacitance-based electrical measurement system to non-invasively distinguish between intact, decayed, and hollow wood based on their respective dielectric responses. In particular, resonance frequency analysis within the 17–30 MHz range proved effective in identifying structural variations, with solid and homogeneous wood samples exhibiting sharper impedance peaks compared to more heterogeneous or degraded specimens. However, the system demonstrated limited sensitivity in distinguishing the severity of decay, suggesting room for improvement in detecting intermediate stages of degradation. To enhance the diagnostic performance of this approach, future studies are encouraged to adopt more controlled experimental conditions and to comprehensively characterize wood properties, such as moisture content, density, and mechanical strength. These efforts would help clarify the extent to which each variable influences the electrical behavior of wood and support more accurate, non-destructive assessment of internal structural integrity.

For Logs 4 to 6, external visual inspection initially suggested that all three were structurally sound, as their outer surfaces appeared intact and free from visible defects. However, upon cross-sectional examination, performed at intervals beginning 15 cm from each end, internal irregularities were detected as early as the first segment.

In Log 4, central degradation was identified in segments L4S1 to L4S3. Although some softwood fibers were still present externally, suggesting partial integrity, the internal condition was compromised. Conversely, segments L4S4 to L4S7 displayed visually sound characteristics, with no obvious signs of internal decomposition, indicating that deterioration was localized toward one end of the log.

Log 5 revealed more extensive internal damage. Segments L5S1 through L5S4 showed significant decay, including localized hollow zones and marked structural compromise. In contrast, segments L5S5 and L5S6 appeared relatively intact, with no visual signs of degradation. Segment L5S7, while structurally intact on visual inspection, exhibited a distinct blackish discoloration, likely indicative of fungal activity. As the sectioning progressed, additional signs of internal cracking and fragmentation emerged, revealing levels of decay not apparent from the external assessment.

In Log 6, small internal voids were present in segments L6S1 through L6S4, suggesting the onset of cavity formation. Segments L6S5 through L6S7, however, showed no visible anomalies or signs of structural failure. These findings underscore the limitations of relying solely on external inspection to evaluate wood quality. The discrepancies between visual appearance and internal condition highlight the importance of employing non-destructive testing methods to assess the true structural integrity of wood.

3.3 Principal Component Analysis and classification

To explore patterns in the data, PCA was applied to the capacitance features. The resulting scatter matrix (Figure 7) displays the distribution of samples across the first four principal components (PC1–PC4). PC1 effectively separates air samples (blue) from all wood categories (Intact, Decay, and Hollow), highlighting the clear dielectric contrast between air and wood. This strong separation confirms that the ECT-S system is highly sensitive to differences in dielectric properties.

Further differentiation is visible in PC2 and PC3, which reveal partial clustering among the wood categories. Hollow (green) and Intact (red) samples begin to diverge in the PC2–PC3 space, whereas Decay (orange) samples overlap with both. This overlap reflects the complex and gradual nature of wood degradation. Notably, the separation of Hollow and Intact samples suggests that differences in material density and dielectric storage capacity are effectively captured through these principal components.

While PC4 contributes less to the total variance and exhibits greater class overlap, its inclusion alongside PC1–PC3 adds to the multidimensional representation of wood structure. Together, these components form a comprehensive framework for clustering and classification tasks, enhancing interpretability in structural diagnosis.

tu_pian_7.png

Figure 7. Principal Component Analysis (PCA) plotting of the capacitance value of wood logs in various conditions

3.4 Classification performance diagnostic reliability

Three classification algorithms, namely SVM, k-NN, and RF, were implemented to assess their ability to differentiate among Intact, Decay, and Hollow wood categories. Each model was trained on 80% of the data and tested on the remaining 20% using stratified sampling to ensure balanced class representation. Input features consisted of the first three principal components (PC1–PC3), which were sufficient to retain the essential structure of the data. Model performance was evaluated through accuracy, precision, recall, and F1-score metrics, supported by confusion matrix analysis. Among the classifiers, the k-NN and RF models demonstrated higher accuracy than SVM (Table 3).

Among the models, SVM yielded the lowest overall accuracy (77%), primarily due to its poor recall for the decay class (0.29). A substantial number of decay samples were misclassified as hollow, indicating difficulty in delineating the subtle transition between partially and severely degraded wood. Nevertheless, SVM performed well in identifying Intact samples with perfect recall and high precision, suggesting that this class presents clearer dielectric separation in the feature space. In contrast, both k-NN and RF achieved higher classification accuracy, each reaching 82%. The k-NN model demonstrated balanced recall across all classes, particularly in detecting decay (0.71) and hollow (0.86). The RF model also performed reliably, with slightly lower consistency between decay and hollow classifications (both with 0.71 recall), but with perfect precision and recall for intact samples. This indicates that while decision boundaries between degraded wood types may remain challenging, ensemble-based classifiers such as RF can still provide robust performance for more well-defined categories. The confusion matrix of k-NN and RF is presented in Figure 8.

Table 3. Summary of classification performance based on Principal Component Analysis (PCA) features

tu_pian_8.png

Figure 8. Confusion matrix of (a) k-NN and (b) Random Forest (RF)

In Figure 8(a), the k-NN model correctly identified 5 out of 7 decay samples, while misclassifying 2 as hollow. Similarly, it correctly classified 6 hollow samples, with one misclassified as decay. All but one intact sample was correctly predicted, indicating strong model sensitivity to undamaged wood, though with a slight overlap toward the decay class.

Figure 8(b) shows the performance of the RF classifier, which also correctly classified 5 decay samples and 5 hollow samples. However, it misclassified two hollow samples as decay, indicating a slight shift in confusion direction compared to k-NN. Notably, all intact samples were classified with perfect accuracy, demonstrating RF’s ability to robustly separate this class. Although the recall for hollow was slightly lower than that of k-NN, the complete absence of errors in the intact category supports the overall strength of ensemble-based classification in identifying clearly defined wood conditions.

3.5 Electrical Capacitance Tomography–Spectroscopy system performance

The flexible capacitive sensor platform, featuring parallel copper electrodes, performed reliably across a range of wood diameters. Its belt-shaped design allowed conformal attachment to wood surfaces, while internal damping mechanisms helped reduce signal noise. The system produced consistent signal responses within the frequency range of 250 kHz to 60 MHz, with distinct resonance behavior typically observed between 17 and 30 MHz.

Despite its effectiveness, the system presented some limitations. The spatial resolution was constrained by the 30 cm segment size, which may limit the detection of small or localized defects. Additionally, signal accuracy could be affected by environmental noise and sensor-to-log contact variability, particularly in field conditions. These issues suggest that improvements in sensor calibration, shielding, and mechanical design are needed to enhance robustness. While the ECT-S system demonstrated high sensitivity for detecting general internal defects, it still faces challenges in resolving intermediate levels of decay, where impedance patterns tend to overlap. The complex interplay of moisture content, density variation, and structural heterogeneity complicates straightforward classification.