1. Introduction
Timber is a commonly used industrial material due to its accessibility and wide variety of mechanical and physical properties, such as higher strength to weight ratio, electrical, and heat insulation and having a simpler and faster installation process [
1]. Timber poles are broadly used worldwide since they are environmental friendly and significantly cheaper than other alternative materials [
1]. Currently, more than 5 million utility timber poles are used in Australia’s infrastructure which represent 80% of the total poles [
2]. Approximately 70% of these timber poles were installed before 1965 [
3]. They will require maintenance and need replacement soon and related cost may reach up to AU
$1.75 billion [
3]. Routine inspections are carried out on a regular basis to monitor pole condition which deteriorates over time due to fungal and termite attacks and climate and soil condition. For maintenance and asset management, utility pole industry spends AU
$50 million yearly to avoid pole failure and ensure reliability of energy network [
3]. In addition, the current pole assessment techniques available in the market mainly rely on visual inspection, sounding, and drilling. However, first two techniques greatly depend on the skill of an asset inspector, which cannot provide consistent accuracy required for ensuring public safety, and drilling is commonly conducted at the ground level reflecting the local condition of the pole. However, damage can present in underground section or top part of the pole which is unreachable by the asset inspector and that often results into inaccurate prediction of the health state. Advanced methods like X-ray, ultrasonic tomography, and radiographic methods use cross-sectional evaluation and can yield highly accurate results. Nevertheless, access to the underground or overhead information remain a challenge. Additionally, use of these methods are limited due to high cost, limited computational efficiency, and complexity in the operation [
4]. To overcome these limitations, variety of non-destructive testing (NDT) methods have been investigated and developed by the engineers and researchers in the past two decades to provide accurate information on the condition of timber pole/pile structures such as wave based, and vibration based, NDT methods.
In case of vibration based method, timber pole is hit manually by a modal hammer and a broadband low frequency input signal is generated [
5]. Because the input frequency is low, generated signal has long wavelength. When the wavelength of the signal gets bigger than the diameter of the pole, the induced stress wave creates a standing wave within the pole resulting into vibration of the whole pole [
6]. Vibration based method has been utilised for non-destructive evaluation of timber poles to assess the underground and top part of the pole since it has higher propagating energies. This method has been successfully used by many researchers for timber poles [
7,
8,
9,
10,
11], tree logs [
12,
13,
14], and pile type structures [
15,
16].
Table 1 demonstrates comparison of the reviewed NDT methods for this study.
In this paper, vibration based method is adopted in order to investigate the condition of utility timber poles since this method can analyse the whole pole by utilizing its dynamic properties such as natural frequency, mode shapes, damping ratio [
17]. Natural frequency is the most popular damage indicator in the early years of vibration-based damage detection as it is easy to determine and is generally less affected by ambient noise [
18,
19,
20,
21,
22]. Natural frequency-based methods depend on the fact that damage in a structure reduces the structural stiffness which actually changes the resonance frequencies. Nonetheless, these methods are model based and need a mathematical model of the structure to obtain vibration frequencies. Another shortcoming is that changes in frequency caused by cracks is not very prominent and is often buried under operational and environmental conditions. Despite some successful application of using natural frequencies as a feature for vibration-based NDT, extensive research proved that natural frequency alone is a poor indicator of damage for the condition assessment due to its low sensitivity to damage especially when severity is low. Hu and Afzal [
10] determined the first two natural frequencies of six damaged timber beams and compared against the same of the intact ones. Their results showed that first natural frequency is not appropriate for identifying low-level damage.
Numerous researchers proposed that mode shapes and their derivatives are more consistent and better indicator than natural frequencies as they have local information that makes them capable to detect multiple cracks. Moreover, they are less effected by environmental change, for instance, temperature [
23]. However, limitations exist in mode shape-based methods. Firstly, a dense array of sensors is required to obtain mode shapes and secondly, they are more susceptible to noise compared to the natural frequencies. Features that utilise natural frequencies and mode shapes together [
24], or derivatives of mode shapes; for example, mode shape curvature [
25], modal flexibility [
26], modal strain energy [
27,
28] are found to be effective for damage identification. Samali et al. [
10] proposed a modified damage index method for locating damage in timber beams by utilising first five mode shapes of the structure and their derivatives attained from the experimental modal analysis. According to the researchers, the modified algorithm is reliable and economical with few anomalies. Hu and Afzal [
23] used a damage indicator derived from the difference between mode shapes with and without damage using discrete Laplace transform for detecting multiple damage in timber beams. The damage indicator was sensitive for designed damage scenarios. However, it was unable to assess the severity of damage quantitatively. Moreover, the method was also computationally and economically expensive, since it required 43 impact points for each set of data in the modal tests. Peterson et al. [
29,
30] adopted a damage index method for a one-dimensional (1-D) system that utilises modal strain energy to detect local damage and decay in timber beams using first two flexural modes. The proposed method found to be effective for single damage scenarios. However, it yielded limited accuracy for the multi damage scenarios.
Nevertheless, these dynamic properties are affected by orthotropic behaviour of timber material, various species of wood used for utility poles, changing length and diameter of poles through the distribution network, difference in climate and soil condition, and presence of additional restraints, transformers, cross-arm, and conductors. Furthermore, environmental factors including temperature and moisture content result in difference of material properties even for the same timber species and geometric location. Additionally, the testing data obtained from practical field are often non-stationary and non-linear since the patterns of acquired data frequently changes with time which means frequency components also changes over time. Consequently, pole assessment results vary ominously from the practical condition of tested poles.
Regarding this problem, advanced signal processing techniques play a significant role in the condition assessment of timber poles. The conventional Fourier transform can be effective; however, it transforms the data from time to frequency domain resulting into complete loss of time information. Due to this deficiency, time–frequency analysis has attracted numerous researchers for analyzing field tested broadband signal. Among all the time frequency analysis techniques, Wavelet Transform (WT) and Wavelet Packet Transform (WPT) are widely adopted by researchers to deal with such signals which yielded acceptable outcomes [
8,
11,
31]. In this technique, original broadband signal was first decomposed into several narrow band signals and then the reconstructed signal was obtained for each band using WT or WPT from which features can be extracted. Usually, resolution and decomposition capability is better for WPT than that of WT while considering a discrete signal [
31]. WT can only be used for decomposing the original signal in low frequency components, whereas WPT can decompose the same signal in both low and high frequency components. As a result, WPT can process a signal in more details than WT [
31]. However, selection of proper mother wavelet is crucial for both WT and WPT. Features that are extracted in vibration based method using signal processing techniques include resonance frequency [
7], power spectral density [
32], continuous wavelet transform coefficient [
8,
9], short kernel coefficient [
9] and auto-regressive coefficient, energy value, and energy coefficient [
11]. The signal processing techniques that are used includes fast Fourier transform [
7], continuous wavelet transform [
9], short kernel method [
9], wavelet packet transform [
11], empirical mode decomposition [
11], and auto-regressive models [
11] in order to extract the above mentioned features. However, these features solely cannot distinguish between serviceable and unserviceable condition poles.
Recently, another time–frequency analysis method named Hilbert–Huang Transform (HHT) has become more popular for analyzing nonstationary and nonlinear signals [
33]. HHT has been adopted for the condition assessment of timber poles using stress wave propagation technique with satisfactory results [
34,
35]. The key part of HHT is empirical mode decomposition (EMD) that can decompose field captured broadband signals into subcomponents or mono-component signals denoted as intrinsic mode function (IMFs). Later, Hilbert transform is applied to those IMFs to obtain instantaneous amplitude and instantaneous frequency, thus Hilbert spectrum can be derived representing full energy–time–frequency distribution. EMD process does not require any convolution [
36]. Compared to the Wavelet transform, the computation time for EMD is also less.
Moreover, it is noteworthy to mention that the concept of the frequency and time resolution is not included in HHT [
37]. Instantaneous frequency obtained from HHT has a meaning only for mono-component signals and signals obtained from field are not mono-component but multicomponent. Hilbert transform of IMFs (generated from EMD process) lead to physically meaningful instantaneous frequencies. However, in practice, HHT suffers from a number of deficiencies. First of all, the EMD process generates some unwanted IMFs in the low frequency range that can be misleading during data interpretation of the data [
36]. Secondly, depending on the analysed signal, the first IMF may cover broad range of frequency in the high frequency range as well which cannot satisfy the mono component property of IMF [
36].
Accordingly, WPT and HHT are employed together in this paper to overcome the aforementioned challenges. This technique is employed on signals associated with three serviceable and three unserviceable timber poles. Selection of mother wavelet is important for the WPT. Maximum energy to Shannon entropy ratio is adopted to select the best mother wavelet function. Using WPT, the captured vibration signal is decomposed into different frequency bands and wavelet coefficients are reconstructed in the dominant frequency bands as new signal for performing further decomposition using EMD. Furthermore, a screening technique is applied to select the dominant IMFs produced from EMD to eliminate the undesirable IMFs by computing the correlation coefficients of the IMFs and the raw analysed signal. Later, Hilbert Transform is applied to those dominant IMFs in order to generate the instantaneous frequency plot.
The effectiveness of the improved HHT is verified for six poles data collected from real field. Sudden changes in the instantaneous frequencies are observed in the analysed signal to determine some vital features related to the health state of timber pole. HHT with the WPT as pre-processor has been successfully used by Z. Peng et al. [
36,
37] for roller bearing fault diagnosis, nevertheless there is still no report on the implementation of this technique in vibration based damage detection of utility timber poles.