Predicting fire resistance ratings of timber structures using artificial neural networks

This paper describes a method to predict the fire resistance ratings of the wooden floor assemblies using Artificial Neural Networks. Experimental data collected from the previously published reports were used to train, validate, and test the proposed ANN model. A series of model configurations were examined using different popular training algorithms to obtain the optimal structure for the model. It is shown that the proposed ANN model can successfully predict the fire resistance ratings of the wooden floor assemblies from the input variables with an average absolute error of four percent.

Journal of Science and Technology in Civil Engineering NUCE 2020 14 (2): 28–39 PREDICTING FIRE RESISTANCE RATINGS OF TIMBER STRUCTURES USING ARTIFICIAL NEURAL NETWORKS Pham Thanh Tunga,∗, Pham Thanh Hungb a Faculty of Building and Industrial Construction, National University of Civil Engineering, 55 Giai Phong road, Hai Ba Trung district, Hanoi, Vietnam b Faculty of Civil Engineering, Hanoi Architectural University, Nguyen Trai road, Hanoi, Vietnam Article history: Received 16/12/2019, Revised 12/01/2020, Accepted 21/01/2020 Abstract This paper describes a method to predict the fire resistance ratings of the wooden floor assemblies using Artificial Neural Networks Experimental data collected from the previously published reports were used to train, validate, and test the proposed ANN model A series of model configurations were examined using different popular training algorithms to obtain the optimal structure for the model It is shown that the proposed ANN model can successfully predict the fire resistance ratings of the wooden floor assemblies from the input variables with an average absolute error of four percent Besides, the sensitivity analysis was conducted to explore the effects of the separate input parameter on the output Results from analysis revealed that the fire resistance ratings are sensitive to the change of Applied Load (ALD) and the number of the Ceiling Finish Layer (CFL) input variables On the other hand, the outputs are less sensitive to a variation of the Joist Type (JTY) parameter Keywords: artificial neural networks; fire resistance; wooden floor assembly; sensitivity analysis https://doi.org/10.31814/stce.nuce2020-14(2)-03 c 2020 National University of Civil Engineering Introduction The ability to maintain the structural integrity of wood structures under fire exposure has been well established Modern buildings with exposed wood structural members are popular since they have a pleasing appearance, easy to use, and offer necessary fire resistance [1] Historically, the height of the conventional wood buildings in the United States was restricted under four stories due to structural barriers and fire concern [2] Thanks to many advanced mechanical properties, the engineered timber products such as Cross-Laminated Timber and Structural Composite Lumber can be used as primary structural materials for the construction of medium-height tall buildings [3] Intensive research has been conducted to enable engineered wood for high-rise buildings in both structural aspects [4–9], as well as fire characteristics [1, 2, 10, 11] Recent research revealed that the fire resistance capacity of the engineered timber, including Glued Laminated Timber and Cross-Laminated Timber, have been proven to outperform that of the lightwood frames and even steel and concrete components [2] Fire performance tests for mass timber had been carried out in Europe [12–15] and recently, in North America [16–18] The tests provided a reliable source to obtain the required minimum fire resistance ratings for structural members ASTM ∗ Corresponding author E-mail address: ptungdhxd@gmail.com (Tung, P T.) 28 Tung, P T., Hung, P T / Journal of Science and Technology in Civil Engineering E119 Standard Test Methods for Fire Tests of Building Construction and Materials [19] or 2015 International Building Code [20] provides the minimum fire resistance requirements for building systems using prescriptive and performance-related provisions Both tested assemblies and methods for calculating fire resistance are provided in the 2015 International Building Codes A Component Additive Method is applied to the building codes to determine the fire resistance ratings of assemblies The method was developed by the National Research Council of Canada in the 1960s It was a result of reviewing the Ten Rules of Fire Endurance Rating [21] for the multiple standard fire test reports A set of rules in the document offers a method to account for the contributions of individual layers to the fire resistance ratings of the assembly Detailed information of these rules is listed in Appendix A The fire endurance ratings of a floor can be estimated either by summing the performance time contribution of (i) the fire-exposed membrane, (ii) framing members, (iii) and any additional protection parts, or performing the standard fire tests For the first method, as stated in the 2015 International Building Code [20] “The fire resistance rating of a wood frame assembly is equal to the sum of the time assigned to the membrane on the fire-exposed side, the time assigned to the framing members and the time assigned for additional contribution by other protective measures such as insulation The membrane on the unexposed side shall not be included in determining the fire resistance of the assembly.” Performance time was assigned for each component of the floor assemblies Table 722.6.2(1) and Table 722.6.2(2) in the 2015 International Building Code presents the time assigned for wallboard membranes and framing members Table shows the time assigned for some popular types of finish materials The time assigned for other members such as wood studs and joists were calculated from ASTM E119 fire resistance tests It worth noting that the fire testing for floor assemblies is normally performed with fire exposure from below, thus the protective membranes on the exposure side would require floor assemblies In addition, the assigned time obtains from membranes for unexposed sides should stand at least 15 minutes Table Time assigned to wall board membranes [20], inch = 2.54 cm Description of finish Time (minutes) 3/8-inch wood structural panel bonded with exterior glue 15/32-inch wood structural panel bonded with exterior glue 19/32-inch wood structural panel bonded with exterior glue 3/8-inch gypsum wallboard 1/2-inch gypsum wallboard 5/8-inch gypsum wallboard 1/2-inch Type X gypsum wallboard 5/8-inch Type X gypsum wallboard Double 3/8-inch gypsum wallboard 1/2-inch + 3/8-inch gypsum wallboard Double 1/2-inch gypsum wallboard 10 15 10 15 30 25 40 25 35 40 An alternative method to estimate the fire resistance ratings of the floor assemblies is to apply Artificial Neural Networks (ANN) The ANN technique can take advantage of the available experimental data and analytical ability of the Artificial Intelligence To perform the ANN method, numerical or 29 Tung, P T., Hung, P T / Journal of Science and Technology in Civil Engineering experimental data collected from the previous publications are used to develop, train, validate, and test ANN models During these processes, the ANN models establish the non-linear relationship between the inputs and the outputs; as a result, the successful ANN models are able to predict the outputs from the unseen input data The ANN method is presented in detail in section of this study Regarding the application of ANN model, a number of research related fire issues are available in the literature For example, Cachim [22] applied the ANN model for calculation of temperatures in timber under fire loading A multilayer feed forward network with three input variables, namely the density of timber, the time of fire exposure, and the distance from the exposed side, were used The output of the model was the temperature in timber The model was trained validated and tested with the numerical data created by numerical simulations Results from the study revealed that the ANN model could accurately calculate the temperature in timber members subjected to fire The application of the ANN model was also found in the research of Tasdemir et al [23] An ANN with four input parameters was used to evaluate the final cross sections of the wooden samples remaining from the fire The experimental tests were also conducted to validate the model A total of 150 experimental test results were used for training and validation of the proposed ANN model, and 30 test results were used for testing The conclusion of the study suggested that the ANN model can be safely used to predict the cross sections of wooden materials remaining from the fire Recently, Naser [24] used ANN models to estimate the thermal and structural properties of timbers at the material and elemental level The study concluded that the method using artificial intelligence could improve the current state of fire resistance evaluation Besides the application for fire-related in wood structures, the ANN model has become a popular technique in many engineering fields For instance, Nguyen and Dinh [25] utilized an ANN model to predict the bridge deck ratings and develop decay curve for the bridge deck In that study, data of 2572 bridges from the National Bridge Inventory were used to develop, train, and test the ANN model The conclusion from the study indicated that the accuracy of bridge rating prediction was 98.5 percent within the margin error of ±1, and the ANN model can effectively be used to develop the bridge deck deterioration curve The ANN model was also used by other investigators for estimating ultimate load carrying of nonlinear inelastic steel truss [26] or predicting the concrete compressive strength [27] The aim of this research is to develop a supervised learning ANN model for predicting the fire resistance ratings of the wooden floor assemblies The proposed ANN model had 11 input variables with one output A number of ANN models with different learning algorithms were developed and evaluated The performance of each model in training, testing, and validation process were compared to acquire the best ANN model Additionally, the selected ANN model was applied to conduct the sensitivity analysis to examine the influence of the input parameters to the output Details of the research are presented in the following sections Data preparation Data used in this research were collected from the previous published technical reports [17, 18], implemented by the National Research Council of Canada The original document contained fire resistance tests results on full-scale floor assemblies of total 85 experimental records Since the experimental tests were conducted on many floor assemblies with various configurations; as a result, some specific parameters in the final reports only contained a limited number of data points In order to obtain the consistent data set, only samples included full records of all parameters were selected In addition, this study focused on wood structures Thus, the floor assemblies with steel joists were removed from the database 30 Tung, P T., Hung, P T / Journal of Science and Technology in Civil Engineering Table Conversion information Type Original values Joist Wood Joist (WJ) Wood-I-Joist (WIJ) Wood Truss (WT) Wood I-Joist flange (WIJ*) Ply Oriented Strand board (OSB) Rock Fiber Insulation Batts (R1) Glass Fiber Insulation Batts (G1) Cellulosic Fiber Insulation (C1) Sub-floor Cavity Insulation Values in Table Table Fire resistance test results Joist Ceiling Finish Type Depth (mm) Spacing (mm) Thickness Layer (mm) JTY JDE JSP CFT 1 1 2 2 2 1 1 3 4 3 1 1 1 4 3 235 235 235 235 240 240 240 240 240 240 235 184 235 235 305 241 305 305 241 241 330 305 286 235 235 235 235 235 235 241 241 305 305 241 305 406 406 406 406 406 406 406 610 610 610 406 406 406 406 406 406 406 610 610 610 406 610 406 406 610 406 610 406 610 406 406 610 610 406 406 12.7 12.7 12.7 12.7 12.7 12.7 12.7 12.7 12.7 12.7 12.7 12.7 15.9 15.9 12.7 15.9 12.7 12.7 12.7 15.9 12.7 12.7 12.7 15.9 12.7 12.7 12.7 12.7 12.7 15.9 15.9 12.7 12.7 15.9 15.9 Sub-Floor Cavity Insulation Type Thickness (mm) Spacing (mm) Applied load (N/m2 ) Fire Resistance Ratings (minutes) SFTH CITY CITH CISP ALD FRR 15.9 15.9 15.9 15.9 15.9 15.9 15.9 19 19 19 15.9 15.5 15.5 15.5 15.5 15.5 15.5 15.5 19 15.5 15.5 19 15.5 15.5 19 15.5 19 15.5 15.5 15.5 15.5 19 19 15.5 15.5 2 1 2 2 1 2 2 2 2 3 1 90 90 90 90 90 90 90 90 90 90 90 89 89 178 89 178 89 89 89 89 89 89 89 89 89 89 89 235 89 241 267 89 89 267 305 406 406 406 406 406 406 406 406 406 610 406 406 203 406 406 406 406 406 406 305 406 610 406 406 610 610 610 610 610 305 305 610 610 305 406 3830 3830 3830 3830 3950 4644 3950 2969 2490 3112 5075 3304 5075 4980 5602 5315 4213 3783 3447 4118 6847 3783 3543 5219 3256 5027 3256 4980 3783 5410 5458 3735 3735 5363 5793 72 67 36 60 64 46 77 75 74 65 65 67 54 59 66 39 68 68 61 50 63 55 64 50 57 57 63 87 59 80 60 56 60 90 99 Type Thickness (mm) CFL SFTY 2 1 2 2 2 1 2 2 2 2 2 2 1 2 2 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 31 235 610 12.7 19 89 610 235 Tung, P T.,12.7 Hung, P T 2/ Journal1of Science and Technology in Civil 406 15.5 89 Engineering 610 3256 57 5027 57 610 that 12.7 values data in19the columns 89 Type, 610 3256Type, and 63 It1 is235 worth the12.7 original of89Joist 235 noting 610 19 610 Sub-floor 3256 57 12.7 22 11 make15.5 235the ANN Cavity Insulation Type not To a readable23input for the values in 235 235 406 406 were 12.7 a number 15.5 89 610610model, 50274980 57 87 these columns were converted into the number The conversion is listed in detail in Table Data after 12.7 22 11 15.5 235 235 610 610 12.7 19 12 8989 610610 32563783 63 59 refinements and conversions are presented in Table The final data consisted of 36 test samples; 15.9 12 11 15.5 241 241 235 406 406 12.7 15.5 33 235 610305 49805410 87 80 each of them included 12 properties The contents from column to column 11 in Table were used 241 235 406 610 12.7 15.5 21 89 610 37835458 59 60 15.9 11 15.5 as the input data for the ANN model,1 and data in column 12 were the 267 output 305 305 241 15.9 12.7 21 11 15.5 19 32 241 89 305610 54103735 80 56 305 241 610 406 15.9 Artificial 12.7 Neural Network 305 610 12.7 21 11 15.5 19 11 267 89 305610 54583735 60 60 19 15.5 89 610 3735 56 90 19 89 610 3735 60 241 610 406 406 3.1 Network 305structure 610 305 406 15.9 12.7 15.9 1 15.5 267 305 305 406 5363 5793 99 Artificial 241 406 Network 15.9 is a 2collection 15.5 neurons 267 grouped 305 in layers, 5363 as depicted 90 An Neural of processing Artificial Neural Network in Fig 1(a) The function of each neuron is to receive input data from connected neurons of the 305 406 15.9 15.5 305 406 5793 99 pre3.1.vious Network structure layer, analysis the data through the weights adjusting procedure, process data (using summation Artificial Neural Network sigmoid functions inathis case),ofand transmits output dataintolayers, the neuron of the subsequent An and Artificial Neural Network is collection processing neurons grouped as depicted in Figure 1a Thelayer function of 3.1 Network structure eachThe neuron is to receive input data from connectedprocessing neurons of the previous analysisintheFig data1(b) through theneurons weights adjusting analyzing scheme of an individual neuron is layer, illustrated The in An Artificial Neural is a collection of sigmoid processing neurons grouped in layers, depicted in Figure 1a to The procedure, processes dataNetwork (using summation and functions in this case), andastransmits output data thefunction neuronofof the each aretoonly connected with neurons from layers.isNo link exists in the eachlayer neuron receive inputscheme data from connected neurons ofother the previous analysis the databetween through theneurons weights adjusting subsequent layer isThe analyzing of an individual processing neuronlayer, illustrated in Figure 1b The neurons in each layer procedure, processes data (using summation and sigmoid functions in this case), and transmits output data to the neuron of the as a same layer The ANN is classified as a shallow network; thus, only three layers of neurons are preare only connected with neurons from other layers No link exists between neurons in the same layer The ANN is classified subsequent layer The analyzing scheme of an individual processing neuron is illustrated in Figure 1b The neurons in each layer shallow network; thus, only three layers of neurons are presented in the ANN structure, namely (i) an input layer, (ii) a hidden sented inconnected the ANN structure, namely (i) anNoinput layer,between (ii) a neurons hiddeninlayer, andlayer (ii)The an ANN output layer The are only with neurons from other layers link exists the same is classified as a layer, and (ii) an output layer The number of neurons in each layer is selected depending on the certain requirements number neurons each depending certain requirements of the problems shallow of network; thus,in only threelayer layersisofselected neurons are presented inon thethe ANN structure, namely (i) an input layer, (ii) a hiddenof the problems layer, and (ii) an output layer The number of neurons in each layer is selected depending on the certain requirements of the problems Inputs Processing Weights Inputs Processing Weights Output Output w x1 x1 w11 xx2 w22 w xx33 w33 w xxnn w wn Sum Sum Sigmoid Sigmoid Out Out n (a) Feed-forward network (b)Individual Individual neuron (b) neuron (b) Individual neuron (a) Feed-forward network (a) Feed-forward network Figure The scheme of an ANN structure 3.2 Performance assessment 3.2 Performance assessment Figure The scheme of an ANN structure Figure The scheme of an ANN structure Performance of the ANN model was evaluated through two factors: coefficient of determination (R2) and Mean Squared of the ANN model was evaluated two factors: of determination (R2equation ) and Mean ErrorPerformance (MSE) The coefficient of determination measures through the correlation betweencoefficient input and output variables using (1) Squared 3.2 Performance assessment Error (MSE) The coefficient of determination measures the correlation between input and output variables using equation (1) / () , ) 𝑅" = − '*01 * +)* // (1) 6,7 * )two () 34 54 ' +) Performance of the ANN model was evaluated through factors: coefficient of determination *01 * 𝑅" = − *01 (1) / 67 34* 54 th Squared Error (MSE) The coefficient th (Rwhere ) andyi isMean of determination be*01𝑦 the i actual output; 𝑦9 is the mean of the actual outputs; outputs; and the n is correlation the total number of ,: is the i predicted measures samples MSE is the mean squared difference between predicted outputs and actual outputs MSE can be calculated using equation tween using (1) outputs; 𝑦,: is the ith predicted outputs; and n is the total number of where yi is input the ith and actualoutput output;variables 𝑦9 is the mean of Eq the actual (2) samples MSE is the mean squared difference between predictedn outputs and actual outputs MSE can be calculated using equation (2) (yi − yˆ i )2 R2 = − i=1 n i=1 32 (1) (yi − y¯ )2 Tung, P T., Hung, P T / Journal of Science and Technology in Civil Engineering where yi is the ith actual output; y¯ is the mean of the actual outputs; yˆ i is the ith predicted outputs; and n is the total number of samples MSE is the mean squared difference between predicted outputs and actual outputs MSE can be calculated using Eq (2) MSE = n n (yi − yˆ i )2 (2) i=1 3.3 Choice of networks Eleven properties of the floor assembly, namely Joist Type (JTY), Joist Depth (JDE), Joist Spacing (JSP), Ceiling Finish Thickness (CFT), Ceiling Finish Layer (CFL), Sub-Floor Type (SFTY), SubFloor Thickness (SFTH), Cavity Insulation Type (CITY), Cavity Insulation Thickness (CITH), Cavity Insulation Spacing (CISP), and Applied Load (ALD), were selected as the input parameters of the ANN model, and the Fire Resistance Ratings (FRR) of the floor assembly was assigned as the output The dataset was divided randomly into three subsets in which 80%, i.e., 26 test samples, of the entire dataset was employed for training model, 10%, i.e., test samples, for validation and the remaining 10%, i.e., test samples, was utilized for testing the prediction accuracy of the ANN model A sigmoid function was selected as an activation function, and the feed-forward back-propagation learning method was assigned for the proposed ANN model The feed-forward back-propagation technique works by using the errors presented in the network output to adjust the weights in each layer in two different processes called feed-forward process and back-propagation process In the feed-forward process the inputs are used to obtain the outputs with some network errors The errors are then passed backwards to the input layers through the back-propagation process, the weights are adjusted during this process to minimize the network errors to an acceptable level To find an optimal training algorithm that works for the available data, eight ANN models were developed and tested with eight popular training algorithms [28] The performances of the models were assessed through MSE values of the four parameters, namely training performance (Train_Perf), testing performance (Test_Perf), validation performance (Validation_Perf), and the number of epochs (Num_Epochs) For each model, the performance result of 10 trials were compared The best performance results from those models are listed in Table It can be seen, the Levenberg-Marquardt algorithm (trainlm) produces the best performance on training, testing, and validation with a low number of epochs For this reason, the Levenberg-Marquardt algorithm was selected for the proposed ANN model Table Performance of the ANN model with different learning algorithms # Algorithm trainrp trainlm traincgp traincgb trainbfg trainoss traincgf traingdx Details Train_Perf Test_Perf Validation_Perf Num_Epochs Resilient Backpropagation Levenberg-Marquardt Polak-Ribiére Conjugate Gradient Conjugate Gradient with Beale Restarts BFGS Quasi-Newton One Step Secant Fletcher-Powell Conjugate Gradient Variable Learning Rate Gradient Descent 33 27.10 0.88 5.98 5.58 16.50 14.30 26.90 25.50 22.20 1.41 6.01 3.04 6.21 2.04 6.83 9.80 11.00 2.46 0.54 3.18 7.89 5.54 12.49 6.24 6 6 6 10 Tung, P T., Hung, P T / Journal of Science and Technology in Civil Engineering To determine the necessary number of nodes in the hidden layer of the proposed ANN model, 20 different ANN models were developed by changing the number of nodes in the hidden layer from one node to 20 nodes Each model was performed ten trials to obtain the average performance results The performance of the ANN models was evaluated through the MSE value of the training, testing, and validation stage with the same dataset Fig presents the performance results from these ANN models The ANN model containing six neurons in the hidden layer generated the best results Consequently, that ANN model was chosen Table presents a brief information of the selected ANN model Table Detailed information of the selected ANN model Parameter Information # neurons in the input layer # neurons in the hidden layer # neurons in the output layer Training method Training algorithm Activation function 11 Feed-forward back-propagation Levenberg-Marquardt (trainlm) Sigmoid Mean Squared Error, mins 150 Training Validation Testing 100 50 0 10 12 14 16 18 20 Number of neurons Model performance of of 20 FigureFigure Model performance 20ANN ANNmodels models Table Detailed information of the selected ANN model Parameter Information Prediction of fire resistance ratings# neurons in the input layer 11 # neurons in the hidden layer 4.1 Applicability of ANN to fire resistance ratings # neurons in theprediction output layer proposedTraining ANN method model back-propagation Performance results of the are presented in Table It isFeed-forward worth noting that the overall performance was calculated for the entire data including training dataset,Levenberg-Marquardt validation dataset Training algorithm (trainlm) and testing dataset As can be seen, the ANN model performed well in all stages with the values of R Activation function Sigmoid were 0.9799, 0.9832, and 0.9778, for training, and testing, respectively Ideally, if a model Prediction of firevalidation, resistance ratings will be equal to The R2 for the overall was 0.9610 perfectly predicts the output, the value of R 4.1 Applicability of ANN to fire resistance ratings prediction2 indicated a good prediction ability of the proposed ANN model Besides R , MSE is an alternative Performance results of the proposed ANN model are presented in Table It is worth noting that the ove indicator that can be used for evaluating of the ANN model Thedataset smaller the MSE calculated forthe the performance entire data including training dataset, validation and testing dataset As can be s performed well in allexperimental stages with the values R2 were 0.9799, 0.9832, and 0.9778, for training, validation, and value is, the stronger the relationship between andofpredicted data For the training data Ideally, if a model perfectly predicts the output, the value of R will be equal to The R2 for the overall w set, the value of MSE was 7.69.good Theprediction MSE values foundANN higher forBesides unseen which indicator were that can be u ability ofwere the proposed model R2, data MSE issets, an alternative performance respectively of the ANN model The smaller the MSE value is, the stronger the relationship between exper 17.7 and 33.1, for testing and validation, data For the training data set, the value of MSE was 7.69 The MSE values were found higher for unseen 17.7 and 33.1, for testing and validation, respectively Table Performance results of ANN model Table Performance results of ANN model Training R2 MSE 0.9799 7.69 The linear regression model The plots for the Validation R 0.9832 MSE 33.1 Training Testing 0.9799 0.9778 7.69 17.7 ValidationOverall Testing 0.9832 33.1 0.9610 0.9778 17.7 12.7 Overall 0.9610 12.7 The linear regression plot was used in this study to present the results from the proposed ANN mo performance of the proposed ANN model at different stages are shown in Figure In these figures, the linea plot was used in between this study to presentresults the results from the proposed ANN the relationship the experimental and the predicted values produced from the model In add shows a perfect correlation between inputs and outputs performance of the proposed ANN model at different stages are shown 34 Tung, P T., Hung, P T / Journal of Science and Technology in Civil Engineering in Fig In these figures, the linear fitting line presents the relationship between the experimental results and the predicted values produced from the model In addition, the “x = y” line shows a perfect correlation between inputs and outputs =0.9799 RR=0.9799 =0.9799 R =0.9799 9090 90 90 8080 8080 7070 7070 6060 6060 5050 5050 4040 4040 3030 3030 3030 3030 4040 4040 5050 5050 6060 6060 7070 7070 =0.9778 RRR =0.9778 =0.9778 R =0.9778 100 100 100 100 Predicteresults results(y), (y),minutes minutes Predicte Predicte results results (y), (y), minutes minutes Predicte Predicte results (y), minutes Predicte Predicteresults results(y), (y),minutes minutes Predicte results (y), minutes 100 100 100 100 Fire resistance ratings Fire resistance ratings Fire resistance ratings Fire resistance ratings Linear fitting Linear fitting Linear fitting Linear fitting x x=x =y= yy x=y 100 8080 9090 100 100 8080 9090 100 9090 90 90 8080 8080 7070 7070 6060 6060 5050 5050 4040 4040 3030 30 3030 30 3030 Experimental results (x), minutes Experimental Experimentalresults results(x), (x),minutes minutes Experimental results (x), minutes 4040 4040 Predicte results(y), (y), minutes Predicte Predicteresults results(y), (y),minutes minutes Predicte results minutes Predicteresults results(y), (y),minutes minutes Predicte Predicte results (y), minutes Predicte results (y), minutes 8080 8080 7070 7070 6060 6060 5050 5050 4040 4040 5050 5050 6060 6060 7070 7070 RRR =0.961 =0.961 =0.961 R =0.961 100 100 100 100 9090 9090 3030 3030 3030 3030 7070 70 70 (b) Validation (b) (b) Validation (b)Validation Validation (b) Validation RR =0.9832 =0.9832 =0.9832 RR=0.9832 4040 4040 6060 60 60 Experimental results (x), minutes Experimental results (x), minutes Experimental results (x), minutes Experimental results (x), minutes Training (a) Training (a) Training (a)(a) Training (a) Training 100 100 100 100 5050 5050 Fire resistance ratings Fire resistance ratings resistance ratings FireFire resistance ratings Linear fitting Linear fitting Linear fitting Linear fitting xx= x= y=yy x=y 100 8080 9090 100 80 80 90 90 100100 Fire Fireresistance resistanceratings ratings Fire resistance ratings Fire resistance ratings Linear Linearfitting fitting Linear fitting Linear fitting x x= =y y x= x= yy 8080 9090 100 100 100 8080 9090 100 9090 9090 8080 8080 7070 7070 6060 6060 5050 5050 4040 4040 3030 30 3030 30 3030 Experimental Experimental results (x), minutes Experimentalresults results(x), (x),minutes minutes Experimental results (x), minutes 4040 4040 5050 5050 6060 60 60 7070 70 70 Fire resistance ratings Fire resistance ratings resistance ratings FireFire resistance ratings Linear fitting Linear fitting Linear fitting Linear fitting x= yy x= x =xy= y 8080 9090 100 100 80 80 90 90 100100 Experimental results (x), minutes Experimental results (x), minutes Experimental results (x), minutes Experimental results (x), minutes (c) (d) (c) Testing (d) Overall Testing (d) Overall (c)Testing Testing (d)Overall Overall (c) (c) Testing (d) Overall Figure Linear regression plot of ANN performance Figure Linear regression plot of ANN performance Figure Figure3.3.Linear Linearregression regressionplot plotofofANN ANNperformance performance The from the model 4a The absolute The experimental data and the predictedvalues values obtained from the ANN model were plotted Figure 4a The absolute The experimental data and the predicted obtained from the ANN model were plotted inin Figure 4a The absolute Theexperimental experimentaldata dataand andthe thepredicted valuesobtained obtained from theANN ANN modelwere wereplotted plottedin inFigure Figure 4a The absolute Figure 3.predicted Linearvalues regression plot ofclear ANN performance prediction errors for each sample were also presented in Figure 4b It is that the proposed ANN model can accurately predict prediction errors for each sample were also presented in Figure 4b It is clear that the proposed ANN model can accurately predict prediction errors for each sample were also presented in Figure 4b It is clear that the proposed ANN model can accurately predict prediction errors for each sample were also presented in Figure 4b It is clear that the proposed ANN model can accurately predict the thefire fire resistance ratings the wooden floor assemblies from the inputs The mean absolute prediction error was about four the fire resistance ratings ofof the wooden floor assemblies from the inputs The mean absolute prediction error was about four the fireresistance resistanceratings ratingsof ofthe thewooden woodenfloor floorassemblies assembliesfrom fromthe theinputs inputs.The Themean meanabsolute absoluteprediction predictionerror errorwas wasabout aboutfour four percent The highest error of about 17 percent was found in test sample number 24, as shown in Figure 4b This can be considered percent The highest error of about 17 percent was found in test sample number 24, as shown in Figure 4b This can be considered The experimental data and the predicted values obtained from the ANN model were plotted in percent bebe considered percent.The Thehighest highesterror errorofofabout about17 17percent percentwas wasfound foundinintest testsample samplenumber number24, 24,asasshown shownininFigure Figure4b 4b.This Thiscan can considered as an issue address this point is database asFig an outliner, and the issue could address this data point excluded from the database asas an outliner, and the issue could address ifif this data point isis excluded from the database 4(a) and The absolute prediction for each sample were also presented in Fig 4(b) It is clear anoutliner, outliner, andthe the issuecould could addressif iferrors thisdata data point isexcluded excludedfrom fromthe the database Predictionerorrs, erorrs,%% Prediction Predictionerorrs, erorrs,% % Prediction FireResistance, Resistance,minutes minutes Fire Resistance, minutes Fire Resistance, minutes Fire 100 100 that the proposed ANN model can accurately predict the18181818fire resistance ratings of the wooden floor 100 100 assemblies from the inputs The mean absolute prediction The highest 1616 error was about four percent X: X: 24 24 1616 9090 X: 16.78 24 24 Y: Y:X:16.78 9090 Y: 16.78 Y: 16.78 error of about 17 percent was found in test sample number 24, as shown in Fig 4(b) This can be 1414 1414 8080 considered as an outliner, and the issue could address if this data point is excluded from the database 8080 1212 1212 7070 7070 1010 1010 4.2 Sensitivity Analysis 6060 6060 88 A sensitivity analysis was performed for the selected8 8ANN model to evaluate the effects of the 66 5050 parameters on the fire resistance ratings In order to6 conduct the sensitivity analysis, each ininput 5050 4040 4040 3030 3030 00 00 55 55 1010 1010 1515 15 2020 20 15 20 Sample Sample### Sample Sample # 2525 2525 Experiment Experiment Experiment Experiment Prediction Prediction Prediction Prediction 3030 3535 3030 3535 (a) (a)Experimental Experimentalvs vsprediction predictionvalues values (a) (a)Experimental Experimentalvsvsprediction predictionvalues values 35 44 4 22 2 00 0000 0 55 5 1010 1010 1515 15 2020 20 15 20 Sample Sample### Sample Sample # 2525 25 25 3030 30 30 (b) (b)Absolute Absoluteprediction predictionerrors errors (b) (b)Absolute Absoluteprediction predictionerrors errors 3535 35 35 (c) Testing (d) Overall Figure Linear regression plotplot of of ANN performance Figure Linear regression ANN performance TheThe experimental datadata andand the the predicted values obtained from thethe ANN model were plotted ininFigure experimental predicted values obtained from ANN model were plotted Figure4a.4a.The Theabsolute absolute prediction errors for each sample were alsoalso presented in Figure 4b.4b It isIt clear that thethe proposed ANN model can prediction errors for each sample were presented in Figure is clear that proposed ANN model canaccurately accuratelypredict predict the the fire fire resistance ratings of the wooden floor assemblies from thethe inputs The mean absolute resistance ratings of the wooden floor assemblies from inputs The mean absoluteprediction predictionerror errorwas wasabout aboutfour four percent TheThe highest error of about 17 percent was found in test sample number 24,24, as as shown in in Figure 4b.4b This percent highest error 17 percent found in test sample number shown Figure Thiscan canbebeconsidered considered Tung,ofP.about T., Hung, P T / was Journal of Science and Technology in Civil Engineering as anasoutliner, andand the the issue could address if this datadata point is excluded from thethe database an outliner, issue could address if this point is excluded from database 100 14 14 80 70 70 60 60 50 50 40 40 Prediction erorrs, % Prediction erorrs, % 80 X: 24 X: 24 Y: 16.78 Y: 16.78 16 90 Fire Resistance, minutes Fire Resistance, minutes 18 16 90 30 18 100 30 12 12 10 10 Experiment Experiment Prediction Prediction 5 10 10 15 15 20 20 25 25 30 30 8 6 4 2 35 35 0 5 10 10 Sample # # Sample 15 15 20 20 25 25 30 30 3535 Sample ## Sample (a) Experimental vs prediction values Absolute prediction errors (a) vs values (b)(b) Absolute prediction (a)Experimental Experimental vsprediction prediction values (b) Absolute prediction errorserrors Figure Performance ANN model Figure Performance of of ANN model Figure Performance of ANN model put parameter was divided into five groups, namely Lowest (Low), Middle Low (Mid-Low), Middle (Mid), Middle High (Mid-High), and Highest (High) [29] The Middle is the mean value of the Lowest and Highest The Middle Low and Middle High represent halfway from the Lowest to the Middle, and from the Middle to the Highest, respectively Detailed values of these input parameters are listed in Table Table Input data for sensitivity analysis Input parameters Low Mid-Low Mid Mid-High High JTY JDE JSP CFT CFL SFTY SFTH CITY CITH CISP ALD FRR 184 406 12.7 15.5 89 203 2490 36 1.75 221 457 13.5 1.25 0.25 16.4 1.50 143 305 3579 52 2.50 257 508 14.3 1.50 0.50 17.3 197 407 4668 68 3.25 294 559 15.1 1.75 0.75 18.1 2.50 251 508 5757 83 330 610 15.9 19 305 610 6847 99 The sensitivity analysis was conducted for each input parameter by changing its value from Low to High while keeping the other inputs constant at the average values The results of the sensitivity analysis for different input parameters are presented in Fig In this figure, the horizontal axis represents the five levels of input variables, while the vertical axis represents the fire resistance ratings of the wooden floor assemblies It can be seen clearly that the fire resistance ratings of the wooden floor assemblies were most 36 FRR 36 52 68 83 99 The sensitivity analysis was conducted for each input parameter by changing its value from Low to High while keeping the other inputs constant at the average values The results of the sensitivity analysis for different input parameters are presented in Figure In this figure, the horizontal axis represents the five levels of input variables, while the vertical axis represents the fire resistance ratings of the wooden floor assemblies Tung, P T., Hung, P T / Journal of Science and Technology in Civil Engineering 80 JTY JDE JSP 70 CFT Fire resistance ratings, mins CFL SFTY SFTH 60 CITY CITH 50 CISP ALD 40 30 20 10 Low Mid-Low Mid Mid-High High Input parameter levels Figure Fire resistance ratings ratings vs Figure Fire resistance vsinputs inputs It can be seen clearly that the fire resistance ratings of the wooden floor assemblies were most sensitive to the Applied Load (ALD) and the number of the Ceiling Finish Layer (CFL) To be specific, for the ALD factor, the fire resistance rating is high when the applied on the floor is low, and vice In the caseof ofthe CFL, an increase in theLayer number(CFL) of ceiling layer would sensitive to theload Applied Load (ALD) andversa the number Ceiling Finish Tofinish be specific, result in an increase of the floor fire resistance capacity By contrast, the Joist Type (JTY) was found to have a minimal for the ALD factor, the fire resistance rating is high when the applied load on the floor is low, andeffect viceon versa In the case of CFL, an increase in the number of ceiling finish layer would result in an increase of the floor fire resistance capacity By contrast, the Joist Type (JTY) was found to have a minimal effect on the fire resistance ratings of the wooden floors In other words, within this study context, a change in the types of joists yielded a limited influence on the fire resistance ratings of the wooden floor assemblies Conclusions In this paper, a method to estimate the fire resistance ratings of the wooden floor assemblies using Artificial Neural Networks was presented A number of ANN models were developed and tested with the experimental data collecting from previous published The selected ANN model performed well in predicting the fire resistance ratings with an average absolute prediction error of about four percent Regarding the sensitivity analysis results, the Applied Load (ALD) and the number of the Ceiling Finish Layer (CFL) input variables were found to have significant effects on the outcome of the ANN model The Joist Type (JTY) parameter, on the other hand, produced an insignificant influence on predicting the output of the proposed ANN model References [1] Technical report No 10 (2015) Calculating the fire resistance of exposed wood members American Wood Council 222 Catoctin Circles, Suit 201 Leesburg, VA 20175 [2] Muszy´nski, L., Gupta, R., hyun Hong, S., Osborn, N., Pickett, B (2019) Fire resistance of unprotected cross-laminated timber (CLT) floor assemblies produced in the USA Fire Safety Journal, 107:126–136 [3] ReThink Wood (2019) Mass timber in North America Retrieved on December 2019 [4] Hummel, J., Seim, W (2019) Displacement-based design approach to evaluate the behaviour factor for multi-storey CLT buildings Engineering Structures, 201:109711 37 Tung, P T., Hung, P T / Journal of Science and Technology in Civil Engineering [5] Nguyen, T T., Dao, T N., Aaleti, S., van de Lindt, J W., Fridley, K J (2018) Seismic assessment of a three-story wood building with an integrated CLT-lightframe system using RTHS Engineering Structures, 167:695–704 [6] Vassallo, D., Follesa, M., Fragiacomo, M (2018) Seismic design of a six-storey CLT building in Italy Engineering Structures, 175:322–338 [7] Nguyen, T T (2017) Modeling of CLT creep behavior and real-time hybrid simulation of a CLT-LiFS building PhD thesis, University of Alabama Libraries [8] Bolvardi, V., Pei, S., van de Lindt, J W., Dolan, J D (2018) Direct displacement design of tall cross laminated timber platform buildings with inter-story isolation Engineering Structures, 167:740–749 [9] Nguyen, T T., Dao, T N., Aaleti, S., Hossain, K., Fridley, K J (2019) Numerical model for creep behavior of axially loaded CLT panels Journal of Structural Engineering, 145(1):04018224 [10] Wiesner, F., Bisby, L (2019) The structural capacity of laminated timber compression elements in fire: A meta-analysis Fire Safety Journal, 107:114–125 [11] Crielaard, R., van de Kuilen, J.-W., Terwel, K., Ravenshorst, G., Steenbakkers, P (2019) Selfextinguishment of cross-laminated timber Fire Safety Journal, 105:244–260 [12] Frangi, A., Fontana, M., Hugi, E., Jăubstl, R (2009) Experimental analysis of cross-laminated timber panels in fire Fire Safety Journal, 44(8):10781087 ă [13] Ostman, B., Mikkola, E., Stein, R., Frangi, A., Kăonig, J., Dhima, D., Hakkarainen, T., Bregulla, J (2010) Fire safety in timber buildings Technical Guideline for Europe [14] Frangi, A., Fontana, M (2010) Fire safety of multistorey timber buildings Proceedings of the Institution of Civil Engineers-Structures and Buildings, 163(4):213–226 [15] Kippel, M., Leyder, C., Frangi, A., Fontana, M (2014) Fire tests on loaded cross-laminated timber wall and floor elements Fire Safety Science, 11:626–639 [16] Dagenais, C (2016) Investigation of fire performance of CLT manufactured with thin laminates FPInnovations, Project 301010618, p 58 [17] Sultan, M A., Seguin, Y P., Leroux, P (1998) Results of fire resistance tests on full-scale floor assemblies Technical report, Institute for Research in Construction [18] Sultan, M A., Latour, J C., Leroux, P., Monette, R C., Seguin, Y P., Henrie, J P (2005) Results of fire resistance tests on full-scale floor assemblies-Phase II Technical report, Institute for Research in Construction [19] ASTM E119-19 (2019) Standard test methods for fire tests of building construction and materials ASTM International, West Conshohocken, PA [20] International Code Council (2015) International building code Falls Church, Va.: International Code Council [21] Harmathy, T Z (1965) Ten rules of fire endurance rating Fire Technology, 1(2):93–102 [22] Cachim, P B (2011) Using artificial neural networks for calculation of temperatures in timber under fire loading Construction and Building Materials, 25(11):4175–4180 [23] Tasdemir, S., Altin, M., Pehlivan, G F., Saritas, I., Erkis, S D B., Tasdemir, S (2015) Determining Fire Resistance of Wooden Construction Elements through Experimental Studies and Artificial Neural Network International Journal of Materials and Metallurgical Engineering, 9(1):209–213 [24] Naser, M Z (2019) Fire resistance evaluation through artificial intelligence-A case for timber structures Fire Safety Journal, 105:1–18 [25] Nguyen, T T., Dinh, K (2019) Prediction of bridge deck condition rating based on artificial neural networks Journal of Science and Technology in Civil Engineering (STCE)-NUCE, 13(3):15–25 [26] Hung, T V., Viet, V Q., Van Thuat, D (2019) A deep learning-based procedure for estimation of ultimate load carrying of steel trusses using advanced analysis Journal of Science and Technology in Civil Engineering (STCE)-NUCE, 13(3):113–123 [27] Naderpour, H., Rafiean, A H., Fakharian, P (2018) Compressive strength prediction of environmentally friendly concrete using artificial neural networks Journal of Building Engineering, 16:213–219 [28] Mathworks (2019) https://www.mathworks.com/help/deeplearning/ug/train-and-apply-multilayerneural-networks.html Retrieved on December 2019 38 Tung, P T., Hung, P T / Journal of Science and Technology in Civil Engineering [29] Nikbin, I M., Rahimi, S., Allahyari, H (2017) A new empirical formula for prediction of fracture energy of concrete based on the artificial neural network Engineering Fracture Mechanics, 186:466–482 Appendix A Harmathy’s Ten Rules of Fire Endurance (Resistance) Rating [21] Rules Contents Explanations The "thermal"fire endurance of a construction consisting of a number of parallel layers is greater than the sum of the "thermal"fire endurance characteristics of the individual layers when exposed separately to fire The fire endurance of a construction does not decrease with the addition of further layers Where two layers of panel materials, such as gypsum wallboard or plywood, are fastened to studs or joists separately, their combined effect is greater than the sum of their individual contributions to the fire endurance rating of the assembly This is a corollary to Rule The fire resistance will not decrease with the addition of layers such as wallboard or other panel materials, regardless of how many layers are added or where they are located within the assembly Wall and ceiling cavities formed by studs and joists protected and encased by wall coverings adds to the fire resistance rating of these assemblies The fire endurance of constructions containing continuous air gaps or cavities is greater than the fire endurance of similar constructions of the same weight but containing no air gaps or cavities The farther an air gap or cavity is located from the exposed surface, the more beneficial its effect on the fire endurance The fire endurance of an assembly cannot be increased by increasing the thickness of a completely enclosed air layer Layers of materials of low thermal conductivity are better utilized on the side of the construction on which fire is more likely to happen The fire endurance of asymmetrical constructions depends on the direction of heat flow The presence of moisture, if it does not result in explosive spalling, increases fire resistance Load-supporting elements, such as beams, girders and joists, yield higher fire endurance when subject to fire endurance tests as parts of floor, roof, or ceiling assemblies than they would when tested separately The load-supporting elements (beams, girders, joists, etc.) of a floor, roof, or ceiling assembly can be replaced by such other load-supporting elements which, when tested separately, yielded fire endurance not less than that of the assembly 10 39 In cases where cavities are formed by joists or studs and protected by 2-inch-thick panel materials against fire exposure, the beneficial effect of such air cavities is greater than if the protection is only 1/2 inch thick An increase in the gap distance between separated layers does not change the fire resistance of an assembly A building material having relatively low thermal conductivity, such as a wood-based material, is more beneficial to the fire resistance of the assembly if placed on the fire-exposed side of the framing than it would be on the opposite side Walls which not have the same panel materials on both faces will demonstrate different fire resistance ratings depending upon which side is exposed to fire This rule results as a consequence of Rules and 6, which point out the importance of location of air gaps or cavities and of the sequence of different layers of solids Materials having a 15 percent moisture content will have greater fire resistance than those having percent moisture content at the time of fire exposure A wood joist performs better when it is incorporated in a floor/ceiling assembly, than tested by itself under the same load A joist in a floor assembly may be replaced by another type of joist having a fire resistance rating not less than that of the assembly  ... minutes Predicte results (y), minutes 100 100 100 100 Fire resistance ratings Fire resistance ratings Fire resistance ratings Fire resistance ratings Linear fitting Linear fitting Linear fitting... (a)(a) Training (a) Training 100 100 100 100 5050 5050 Fire resistance ratings Fire resistance ratings resistance ratings FireFire resistance ratings Linear fitting Linear fitting Linear fitting... xx= x= y=yy x=y 100 8080 9090 100 80 80 90 90 100100 Fire Fireresistance resistanceratings ratings Fire resistance ratings Fire resistance ratings Linear Linearfitting fitting Linear fitting Linear