摘要: 目的我国有大量的木结构古建筑，在现场对古建筑木构件正常木材的物理力学性能给予方便的检测评估，是古建筑木结构日常保护、修缮和安全评估的刚性需求。本研究对敲击声信号引入机器学习算法处理，力图将便捷的敲击方式应用于古建筑木材物理力学性能的无损检测、�/sec> 方法以北京某皇家古建筑拆修下来的4段落叶松旧木构件为原材料，加工无疵试件，首先探究木试件尺寸、密度对敲击声信号的影响，试验测定木试件的密度、抗弯强度、抗弯弹性模量、顺纹抗压强度等物理力学性能参数；然后对试验采集的敲击声信号进行梅尔频率谱系数（MFSC）特征提取，以敲击声MFSC特征为输入、试件物理力学性能为输出，构建古建筑木材物理力学性能卷积神经网络（CNN）评估模型、�/sec> 结果试件尺寸对敲击声信号没有影响，密度较高试件的敲击声信号的主峰频率较高；失活层对模型性能有较为明显的影响，失活层失活率为0.2时的拟合效果最佳；所建立的模型对古建筑木材物理力学性能的评估效果良好，密度、抗弯强度、抗弯弹性模量、顺纹抗压强度评估值与真实值之间的决定系数分别达到0.873�?.819�?.746�?.860、�/sec> 结论本研究构建的基于敲击声MFSC特征CNN模型，对古建筑木材物理力学性能进行检测评估是可行的、�/sec>

关键诌�
• 古建筑木构件/
• 物理力学性能/
• 敲击�?/a> /
• 梅尔频率谱系数（MFSC（�/a> /
• 卷积神经网络（CNN（�/a>

Abstract: ObjectiveThere are a large number of ancient buildings with wooden structures in China. How to conveniently test and evaluate the physical and mechanical properties of ancient building normal timber is a rigid requirement for the daily protection, repair and safety assessment of ancient wood building structures. In this study, the machine learning algorithm was introduced to the knocking sound signal, and try to apply the convenient knocking method to the nondestructive testing of the physical and mechanical properties of ancient building wood. MethodThe four-section ancient larch ( Larix gmelinii) timber members dismantled from an ancient royal building in Beijing were used as raw materials to process clear specimens. Firstly, the influence of size and density of the wood specimen on the knocking sound signal was investigated. And the physical and mechanical property parameters such as density, modulus of rupture, modulus of elasticity and compressive strength parallel to grain of wood specimens were measured experimentally. Then, the Mel frequency spectral coefficient (MFSC) feature extraction was carried out on the knocking sound signal collected in the experiment. Taking the knocking sound MFSC feature as the input, and the physical and mechanical properties of the specimen as the output, a convolutional neural network (CNN) evaluating model for the physical and mechanical properties of ancient building timber member was constructed. ResultThe size of the specimen had no effect on the knocking signal, and the dominant peak frequency of the knocking signal of the specimen with higher density was higher. The dropout layer had a significant impact on the performance of the model, and the fitting effect was the best when the dropout rate was 0.2. The established model had a good effect on the evaluation of physical and mechanical properties of ancient building wood, and the coefficient of determination between the evaluation value of density, modulus of rupture, modulus of elasticity and compressive strength parallel to grain and the real value reached 0.873, 0.819, 0.746 and 0.860, respectively. ConclusionThe CNN model constructed in this study based on the MFSC feature of knocking sound is feasible to detect and evaluate the physical and mechanical properties of timber in ancient buildings.

Key words:
• ancient building timber member/
• physical and mechanical property/
• knocking sound/
• Mel-frequency spectral coefficient (MFSC)/
• convolutional neural network (CNN)

�?nbsp; 1敲击检测原琅�/p>

k_A咋�i>k_B分别表示木构件低密度和高密度的局部结构刚度、�i>k_Aandk_Brepresent the local structural stiffness of low density and high density wood members, respectively.

Figure 1.Knocking test theory

下载: 全尺寸图牆�/a> 幻灯牆�/a>

�?nbsp; 2MFSC特征提取流程

Figure 2.MFSC feature extraction process

下载: 全尺寸图牆�/a> 幻灯牆�/a>

�?nbsp; 3原材料与试验试件

Figure 3.Raw material and experimental samples

下载: 全尺寸图牆�/a> 幻灯牆�/a>

�?nbsp; 4敲击试验装置

Figure 4.Knocking test device

下载: 全尺寸图牆�/a> 幻灯牆�/a>

�?nbsp; 5卷积神经网络结构国�/p>

Figure 5.Structure diagram of CNN

下载: 全尺寸图牆�/a> 幻灯牆�/a>

�?nbsp; 6密度和力学性能的分布图

ρ. 密度 Density; MOR. 抗弯强度 Modulus of rupture; MOE. 抗弯弹性模� Modulus of elasticity; CSPG. 顺纹抗压强度 Compressive strength parallel to grain

Figure 6.Distribution map of density and mechanical properties

下载: 全尺寸图牆�/a> 幻灯牆�/a>

�?nbsp; 7密度与力学性能参数的相关�?/p>

Figure 7.Correlations between density and mechanical property parameters

下载: 全尺寸图牆�/a> 幻灯牆�/a>

�?nbsp; 8采集的敲击声音信叶�/p>

Figure 8.Collected knocking sound signal

下载: 全尺寸图牆�/a> 幻灯牆�/a>

�?nbsp; 9密度对主峰频率的影响

Figure 9.Effects of specimen density on the dominant peak frequency

下载: 全尺寸图牆�/a> 幻灯牆�/a>

�?nbsp; 10声音信号的梅尔频率谱系数(MFSC)特征

Figure 10.Mel frequency spectral coefficients (MFSC) feature of sound signal

下载: 全尺寸图牆�/a> 幻灯牆�/a>

�?nbsp; 11失活层参数对结果的影哌�/p>

Figure 11.Influence of dropout layer parameters on results

下载: 全尺寸图牆�/a> 幻灯牆�/a>

�?nbsp; 12测试值与评估值间的关糺�/p>

M_APE为绝对百分比误差的平均值、�i>M_APEis the value of mean absolute percentage error.

Figure 12.Relationship between experimental values and evaluating values

下载: 全尺寸图牆�/a> 幻灯牆�/a>

�?nbsp; 1试验试件

Table 1.Test specimens

尺寸 Size	数量 Number	用� Application
60 mm × 60 mm × 60 mm	4	探究试件尺寸、密度对敲击声音信号的影� Exploring the influence of specimen size and density on knocking sound signal
40 mm × 40 mm × 40 mm	4
20 mm × 20 mm × 20 mm	4
20 mm × 20 mm × 20 mm	65	敲击试验 Knocking test
20 mm × 20 mm × 300 mm	65	物理力学性能测定 Determining the physical and mechanical properties

下载: 导出CSV

�?nbsp; 2CNN网络模型结构参数

Table 2.Structural parameters of CNN network model

类别 Category	层数 Layer number	输出参数 Output parameter	步长 Step size	卷积核尺� Kernel size
输入� Input layer	1	�?,35,40（�/td>
卷积� Convolutional layer	2	�?2,33,38（�/td>	1	32 × 3 × 3
最大池化层 Max. pooling layer	3	�?2,16,19（�/td>	2	2 × 2
卷积� Convolutional layer	4	�?4,14,17（�/td>	1	64 × 3 × 3
最大池化层 Max. pooling layer	5	�?4,7,8（�/td>	2	2 × 2
失活� Dropout layer	6	�?2,7,8（�/td>
全连接层 Full connect (FC) layer	7	�?,512（�/td>
输出� Output layer	8	�?（�/td>

下载: 导出CSV

�?nbsp; 3试件密度和力学性能参数

Table 3.Density and mechanical characteristic parameters of specimens

统计� Statistic	MOE/MPa	MOR/MPa	CSPG/MPa	ρ/（kg·m�?（�/td>
AV	5 517.25	64.30	41.78	484.05
SD	1 602.23	19.42	7.92	101.07
CV	0.29	0.30	0.19	0.21
注： AV表示平均值，SD表示标准差，CV表示变异系数。下同� Notes: AV represents average value; SD represents the standard deviation; CV represents the coefficient of variation. The same below.

下载: 导出CSV

�?nbsp; 4不同尺寸下声音信号的主峰频率

Table 4.Dominant peak frequency of sound signal at different sizes

编号 No.	密度 Density/（kg·m�?（�/td>	主峰频率 Dominant peak frequency/Hz
		60 mm × 60 mm × 60 mm	40 mm × 40 mm × 40 mm	20 mm × 20 mm × 20 mm	AV	SD	CV
试件1 Sample 1	400.78	370.00	370.00	370.00	370.00	0	0
试件2 Sample 2	446.65	395.55	395.50	385.45	392.17	5.18	1.32%
试件3 Sample 3	564.39	466.00	467.33	471.10	468.14	2.65	0.57%
试件4 Sample 4	623.57	592.50	590.00	592.50	591.67	1.45	0.24%

下载: 导出CSV

[2]Patrícia C R, Michael A, José A C, et al. Non-destructive structural wood diagnosis of a medieval building—sciencedirect[J]. Procedia Structural Integrity, 2017, 5: 1147�?152. doi:10.1016/j.prostr.2017.07.024 [3]Tallavo F, Cascante G, Pandey M D, et al. A novel methodology for condition assessment of wood poles using ultrasonic testing[J]. NDT & E International, 2012, 52: 149�?56. [4]Perlin L P, Pinto R C, Valle A D. Ultrasonic tomography in wood with anisotropy consideration[J]. Construction and Building Materials, 2019, 229: 116958. doi:10.1016/j.conbuildmat.2019.116958 [5]于永�? 管成, 张厚�? �? 古建筑墙体木柱缺陷对其安全性影响数值模拟研究[J]. 北京林业大学学报, 2022, 44(1): 132�?45. doi:10.12171/j.1000-1522.20210341

Yu Y Z, Guan C, Zhang H J, et al. Numerical simulation on the influence of wall wood column defects on the safety of ancient building[J]. Journal of Beijing Forestry University, 2022, 44(1): 132�?45. doi:10.12171/j.1000-1522.20210341 [6]陈九�? 陈雪�? 戴璐. 残损斗栱节点受力性能试验研究[J]. 北京林业大学学报, 2020, 42(2): 149�?58. doi:10.12171/j.1000-1522.20190278

Chen J Z, Chen X Y, Dai L. Experimental research on mechanical performance of damaged bracket set joints[J]. Journal of Beijing Forestry University, 2020, 42(2): 149�?58. doi:10.12171/j.1000-1522.20190278 [7]Feio A, Machado J S. In-situ assessment of timber structural members: combining information from visual strength grading and NDT/SDT methods: a review[J]. Construction and Building Materials, 2015, 101: 1157�?165. doi:10.1016/j.conbuildmat.2015.05.123 [8]Jasieńko J, Nowak T, Hamrol K. Selected methods of diagnosis of historic timber structures: principles and possibilities of assessment[J]. Advanced Materials Research, 2013, 778: 225�?32. doi:10.4028/www.scientific.net/AMR.778.225 [9]Íñiguez-González G, Arriaga F, Esteban M, et al. Reference conditions and modification factors for the standardization of nondestructive variables used in the evaluation of existing timber structures[J]. Construction and Building Materials, 2015, 101: 1166�?171. doi:10.1016/j.conbuildmat.2015.05.128 [10]张厚�? 管成, 文剑. 木质材料无损检测的应用与研究进展[J]. 林业工程学报, 2016, 1(6): 1�?.

Zhang H J, Guan C, Wen J. Applications and research development of nondestructive testing of wood based materials[J]. Journal of Forestry Engineering, 2016, 1(6): 1�?. [11]孙燕�? 张厚�? 朱磊, �? 木构件材料力学性能快速检测研究[J]. 西北林学院学�? 2012, 27(2): 245�?48. doi:10.3969/j.issn.1001-7461.2012.02.50

Sun Y L, Zhang H J, Zhu L, et al. Rapid test on mechanical properties of wooden components[J]. Journal of Northwest Forestry University, 2012, 27(2): 245�?48. doi:10.3969/j.issn.1001-7461.2012.02.50 [12]孙燕�? 张厚�? 朱磊, �? 基于微钻阻力的落叶松弹性模量快速检测[J]. 湖北农业科学, 2012, 51(11): 2348�?350. doi:10.3969/j.issn.0439-8114.2012.11.055

Sun Y L, Zhang H J, Zhu L, et al. Research on rapid detection of larch wood modulus of elasticity based on micro-drilling resistance[J]. Hubei Agricultural Sciences, 2012, 51(11): 2348�?350. doi:10.3969/j.issn.0439-8114.2012.11.055 [13]朱磊, 张厚�? 孙燕�? �? 基于应力波和微钻阻力的红松类木构件力学性能的无损检测[J]. 南京林业大学学报(自然科学�?, 2013, 37(2): 156�?58.

Zhu L, Zhang H J, Sun Y L, et al. Mechanical properties non-destructive testing of wooden components of Korean pine based on stress wave and micro-drilling resistance[J]. Journal of Nanjing Forestry University (Natural Science Edition), 2013, 37(2): 156�?58. [14]朱磊, 张厚�? 孙燕�? �? 基于应力波和微钻阻力的古建筑木构件材料力学性能检测[J]. 东北林业大学学报, 2011, 39(10): 81�?3. doi:10.3969/j.issn.1000-5382.2011.10.023

Zhu L, Zhang H J, Sun Y L, et al. Determination of mechanical properties of ancient architectural timber based on stress wave andmicro-drilling[J]. Journal of Northeast Forestry University, 2011, 39(10): 81�?3. doi:10.3969/j.issn.1000-5382.2011.10.023 [15]Cascón I, Sarasua J A, Tena M, et al. Development of an acoustic method for wood disease assessment[J]. Computers and Electronics in Agriculture, 2021, 186(5): 106195. [16]邬冠�? 林俊�? 任吉�? �? 声振检测方法的发展[J]. 无损检�? 2011, 33(2): 35�?1.

Wu G H, Lin J M, Ren J L, et al. Evolution of the acoustic impact testing method[J]. Nondestructive Testing, 2011, 33(2): 35�?1. [17]Senalik C A, Zhou L, Ross R J. Assessment of deterioration in timbers with time and frequency domain analysis techniques[C]//Wang X, Senalik C A, Ross R J .20th International Nondestructive Testing and Evaluation of Wood Symposium. Madison: Department of Agriculture, Forest Service, Forest Products Laboratory, 2017: 298�?06. [18]Xin Z, Zhang H, Guan C, et al. Screening for internal defects of timber members based on impact evaluation method[J]. Construction and Building Materials, 2020, 263(10): 120145. [19]Esteban L G, Fernández F G, Palacios P D. Prediction of plywood bonding quality using an artificial neural network[J]. Holzforschung, 2011, 65(2): 209�?14. [20]Nasir V, Nasir V, Fathi H, et al. Prediction of mechanical properties of artificially weathered wood by color change and machine learning[J]. Materials, 2021, 14(21): 6314. doi:10.3390/ma14216314 [21]Xin Z B, Ke D F, Zhang H J, et al. Non-destructive evaluating the density and mechanical properties of ancient timber members based on machine learning approach[J]. Construction & Building Materials, 2022, 341: 127855. [22]Yuan C, Zhang J, Chen L, et al. Timber moisture detection using wavelet packet decomposition and convolutional neural network[J]. Smart Materials and Structures, 2021, 30(3): 1�?3. [23]Cawley P, Adams R D. The mechanics of the coin-tap method of non-destructive testing[J]. Journal of Sound and Vibration, 1988, 122(2): 299�?16. doi:10.1016/S0022-460X(88)80356-0 [24]Cawley P. Low frequency NDT techniques for the detection of disbonds and delaminations[J]. Ndt & E International, 1992, 25(2): 100. [25]Volkmann J, Stevens S S, Newman E B. A scale for the measurement of the psychological magnitude pitch[J]. Journal of the Acoustical Society of America, 1937, 8(3): 185�?90. doi:10.1121/1.1915893 [26]张钰�? 蒋盛�? 基于MFCC特征提取和改进SVM的语音情感数据挖掘分类识别方法研究[J]. 计算机应用与软件, 2020, 37(8): 160�?65. doi:10.3969/j.issn.1000-386x.2020.08.028

Zhang Y S, Jiang S Y. Speech emotion data mining classification and recognition method based on MFCC feature extraction and improved SVM[J]. Computer Applictions and Software, 2020, 37(8): 160�?65. doi:10.3969/j.issn.1000-386x.2020.08.028 [27]中国林业科学研究院木材工业研究所, 中国国家标准化管理委员会. 木材含水率测定方泔� GB/T 1931�?009[S]. 北京: 中国标准出版�? 2009: 8.

Reasearch Insitute of Wood Industry, Standardization Administration. Method for determination of the moisture content of wood: GB/T 1931?2009[S]. Beijing: Standard Press of China, 2009: 8. [28]中国林业科学研究院木材工业研究所, 中国家标准化管理委员�? 木材抗弯弹性模量测定方泔� GB/T 1936.2�?009[S]. 北京: 中国标准出版�? 2009: 8.

Reasearch Insitute of Wood Industry, Standardization Administratio. Method for determination of the modulus of elasticity in static bending of wood: GB/T 1936.2?2009[S]. Beijing: Standard Press of China, 2009: 8. [29]中国林业科学研究院木材工业研究所, 中国国家标准化管理委员会. 木材抗弯强度试验方法: GB/T 1936.1�?009[S]. 北京: 中国标准出版�? 2009: 8.

Reasearch Insitute of Wood Industry, Standardization Administration. Method of testing in bending strength of wood: GB/T 1936.1?2009[S]. Beijing: Standard Press of China, 2009: 8. [30]中国林业科学研究院木材工业研究所, 中国国家标准化管理委员会. 木材顺纹抗压强度试验方法: GB/T 1935�?009[S]. 北京: 中国标准出版�? 2009: 8.

Reasearch Insitute of Wood Industry, Standardization Administration. Method of testing in compressive strength parallel to grain of wood: GB/T 1935?2009[S]. Beijing: Standard Press of China, 2009: 8. [31]中国林业科学研究院木材工业研究所, 中国国家标准化管理委员会. 木材密度测定方法: GB/T 1933�?009[S]. 北京: 中国标准出版�? 2009: 8.

Reasearch Insitute of Wood Industry, Standardization Administration. Method for determination of the density of wood: GB/T 1933�?009[S]. Beijing: Standard Press of China, 2009: 8. [32]Yokoyama M, Gril J, Matsuo M, et al. Mechanical characteristics of aged Hinoki wood from Japanese historical buildings[J]. Comptes Rendus Physique, 2009, 10(7): 601�?11. doi:10.1016/j.crhy.2009.08.009 [33]孙燕�? 基于微钻阻力的古建筑木材密度与力学性能检测研究[D]. 北京: 北京林业大学, 2012.

Sun Y L. Determining wood density and mechanical properties of ancient architectural timbers with micro-drilling resistance[D]. Beijing: Beijing Forestry University, 2012. [34]朱磊. 基于应力波的古建筑木构件材料力学性能检测技术研究[D]. 北京: 北京林业大学, 2012.

Zhu L. Determining the mechanical properties of ancient architectural timber with stress waves[D]. Beijing: Beijing Forestry University, 2012.