Intelligent composite fastener, manufacturing process and health monitoring method

Abstract

An intelligent composite fastener, manufacturing process and health monitoring method, the thread of the fastener is coated with a nano coating, and the shank of the fastener is embedded with a neodymium iron boron magnet. Based on the magnetic-force effect, a physically constrained guided neural network (PGNN) is proposed to establish the mapping between the magnetic induction intensity and the bearing stress during loading. Due to the strengthening effect of the nano coating, the strength of the composite joint structure is strengthened, and its bearing stress can reach 196±12MPa, which is 26.5% higher than that of the composite joint structure using uncoated composite fasteners. When the full composite joint structure fails, the nano-coated composite fastener is sheared at the screw head, while the uncoated composite fastener is sheared at the thread. The bearing stress predicted by PGNN can reasonably quantify the actual stress of the test joint. Through experiments, it is found that the maximum prediction error of PGNN is only 18.9MPa, which is reduced by 51.7% compared with the neural network (NN) without physical constraint guidance.

Description

Technical Field

[0001] This invention relates to the field of composite material technology, and in particular to an intelligent composite material fastener, manufacturing process, and health monitoring method. Background Technology

[0002] Structural health monitoring has been applied in engineering in various ways. Currently, sensors are widely used, such as piezoelectric sensors, piezoresistive sensors, and grating sensors. Most of these sensors rely on current conduction, and some functional components are unavoidable; therefore, these sensors require wiring and are quite bulky. This paper presents a new approach: based on the magneto-force effect of ferromagnetic materials, using a small, wireless sensor for structural health monitoring.

[0003] Therefore, this study proposes a smart composite fastener with a nano-coating on the thread surface and a wireless neodymium iron boron magnet embedded in the screw. The nano-coating, composed of nanofillers (graphite, MoS2, Al2O3, CeF3), polyimide, and dimethylformamide, can repair machining defects in the thread and improve the hardness and wear resistance of the thread surface. Based on the magneto-force effect, the changes in load-bearing stress and magnetic induction intensity are recorded during loading, and a physically constrained guided neural network is used to establish a mapping relationship between magnetic induction intensity and load-bearing stress. This study provides a method for weight reduction and structural health monitoring of composite structures.

[0004] Due to their superior mechanical properties and weight reduction advantages, carbon fiber reinforced polymer matrix composites are increasingly used in aircraft main structures. During assembly, composite structures are often connected using metal mechanical fasteners. However, the use of metal fasteners diminishes the weight reduction advantages of composite structures, and electrochemical corrosion between carbon fibers and metals is also a problem. Furthermore, metal fasteners can act as an ignition source, potentially causing catastrophic consequences if the aircraft is struck by lightning. Another problem with composite connection structures formed using metal fasteners is the difficulty in inspecting the joint areas; a sudden failure in any part could be fatal to the aircraft. Therefore, regular inspection of structural health, especially at joints, is essential.

[0005] To avoid the problems caused by metal fasteners, composite fasteners have emerged as a potential alternative. Some researchers have machined threads into composite fasteners and found that the strength of all-composite connection structures is not ideal. This is because machining introduces defects into the thread surface, and stress concentration occurs at the threads under load, causing the threads to fail earlier than the screw head, thus failing to fully utilize the load-bearing capacity of the composite fastener. Summary of the Invention

[0006] To address the problems in the prior art, this application proposes a smart composite material fastener, wherein the threads of the fastener are coated with a nano-coating and neodymium iron boron magnets are embedded in the shaft of the fastener.

[0007] Preferably, the nano-coating comprises polyimide, dimethylformamide, and functional nanofillers; the functional nanofillers include nano-graphite, nano-MoS2, nano-Al2O3, and nano-CeF3.

[0008] Preferably, dimethylformamide is completely dissolved in polyimide to form a solution, and the functional nanofiller is stirred and mixed with the solution using a magnetic stirrer to form a mixture; the composite fastener and nut are immersed in the mixture and cured at high temperature; after embedding neodymium iron boron magnets into the composite fastener, it is sealed with sealant.

[0009] Preferably, a mapping between the magnetic induction intensity of the smart composite fastener and the load-bearing stress during loading is established based on the magnetic-force effect and a neural network containing physical constraints; the neural network containing physical constraints is trained to predict the load-bearing stress of the connection structure and monitor the structural health.

[0010] Preferably, the neural network includes an input layer, a hidden layer, and an output layer, and the training process includes:

[0011] Multiple tensile tests were conducted on the composite material structure formed by the nano-coated composite fastener.

[0012] Before failure, multiple data points were selected from the magnetic flux density curve and stress-strain curve to form multiple sets of input and output data, where the input data is the normal magnetic flux density M. x and tangential magnetic induction intensity M y The output data is the bearing stress S;

[0013] The neural network is trained and tested using multiple sets of input and output data.

[0014] Preferably, the physical constraint equation is expressed in the form of the derivatives of the bearing stress S with respect to the tangential magnetic induction intensity My and the normal magnetic induction intensity Mx:

[0015]

[0016] By weight Weight constraints are added to incorporate into the neural network; each constraint is either non-positive or non-negative to ensure that the bearing stress S is relative to the tangential magnetic induction intensity M. y and normal magnetic induction intensity M x The derivatives of f1 and f2 are non-positive or non-negative; the derivatives of activation functions f1 and f2 are non-negative, and the weight constraints are:

[0017]

[0018] Preferably, the mean squared error loss function of the neural network is expressed as:

[0019]

[0020] Where S i It is predicted data, S it This is the test data, and n is the sample size.

[0021] Preferably, the mathematical expression for the hidden layer neurons is:

[0022]

[0023]

[0024] The mathematical expression for the output layer neuron is:

[0025]

[0026]

[0027] in, This represents the weight of the j-th neuron in the L-th layer. This represents the bias of the j-th neuron in the L-th layer. This represents the linear result of the j-th neuron in the L-th layer. This represents the activation output of the j-th neuron in layer L.

[0028] The above-mentioned technical features can be combined in various suitable ways or replaced by equivalent technical features, as long as the purpose of the present invention can be achieved.

[0029] The intelligent composite material fastener, manufacturing process, and health monitoring method provided by this invention have at least the following advantages compared with the prior art:

[0030] 1. The nano-coating enhances the strength of the composite material connection structure and improves the hardness and wear resistance of the threads;

[0031] 2. It achieved the reduction of structural mass and the monitoring of structural health status;

[0032] 3. Improved the consistency of neural network prediction results and reduced the maximum prediction error of neural networks. Attached Figure Description

[0033] The invention will now be described in more detail with reference to embodiments and the accompanying drawings.

[0034] Figure 1The image shows a comparison of the surfaces of threads without nano-coating and threads with nano-coating.

[0035] Figure 2 This diagram illustrates the health monitoring principle based on the magneto-force effect of the present invention.

[0036] Figure 3 A schematic diagram of the static tensile test of the present invention is shown;

[0037] Figure 4 The PGNN structure diagram of the present invention is shown;

[0038] Figure 5(a) shows the stress-strain curves and load-bearing strength diagrams of the NSCF connection structure and SCF connection structure of the present invention;

[0039] Figure 5(b) shows the curves of the change in normal and tangential magnetic induction intensity of the NSCF connection structure during the loading process of the present invention;

[0040] Figure 5(c) shows the failure modes of the NSCF connection structure and SCF connection structure of the present invention.

[0041] Figure 6(a) shows the predictive randomness of the neural network guided by physical constraints of the present invention;

[0042] Figure 6(b) shows the prediction accuracy of the physical constraint-guided neural network of the present invention;

[0043] Figure 6(c) shows the loss values ​​of PGNN and NN during neural network training;

[0044] Figure 7 A schematic diagram of the short-cut carbon fiber processing procedure of the present invention is shown;

[0045] Figure 8 A schematic diagram of the preparation process of the nano-coated thread of the present invention is shown. Detailed Implementation

[0046] The invention will now be further described with reference to the accompanying drawings.

[0047] This invention provides a smart composite fastener with a nano-coating on its threads and a neodymium iron boron magnet embedded in its shank. Based on the magneto-force effect, a physically constrained guided neural network (PGNN) is proposed to establish a mapping between magnetic induction intensity and load-bearing stress during loading. Due to the strengthening effect of the nano-coating, the strength of the composite connection structure is enhanced, and its load-bearing stress can reach 196±12 MPa, which is 26.5% higher than that of a composite connection structure using uncoated composite fasteners. When the all-composite connection structure fails, the nano-coated composite fastener shears at the screw head, while the uncoated composite fastener shears at the thread. The load-bearing stress predicted by PGNN can reasonably quantify the actual stress of the test joint. Experiments show that the maximum prediction error of PGNN is only 18.9 MPa, which is 51.7% less than that of a neural network (NN) without physical constraints. Furthermore, compared with NN, PGNN exhibits lower prediction randomness, better convergence, and higher prediction accuracy. This study provides a potential alternative to metal fasteners, and by combining a wireless load-bearing stress prediction model, it achieves the goals of reducing structural mass and monitoring structural health.

[0048] In one embodiment, to improve the hardness and wear resistance of the threads, this application designs a nano-coating. The nano-coating consists of a solvent, a binder, and functional nanofillers. Polyimide (PI) is used as the solvent, dimethylformamide (DMF) as the binder, nano-graphite (8 μm) and nano-MoS2 (2 μm) are used to improve lubricity, nano-Al2O3 (20 μm) is used to improve hardness, and nano-CeF3 (3 μm) is used to improve wear resistance.

[0049] like Figure 1 As shown, compared to uncoated threads, the threads of nano-coated composite fasteners have a smooth surface with almost no visible defects. To monitor the health of the structure, such as... Figure 2 As shown, a neodymium iron boron magnet is embedded in a nano-coated composite fastener. Based on the magneto-force effect, changes in magnetic flux density are recorded to predict structural stress. Normal and tangential magnetic flux densities are measured using a CH-3600 magnetic sensor.

[0050] In one embodiment, a quasi-static tensile test is performed using a testing machine, such as... Figure 3 As shown, the tensile speed was set to 1 mm / min. The composite fasteners were tightened using a torque wrench with a tightening torque of 8.50 N·m. To obtain training and testing data, 40 tensile tests were conducted on the composite structure formed by the nano-coated composite fasteners. For comparison, 5 tensile tests were conducted on the composite connection structure formed by the uncoated composite fasteners.

[0051] In one embodiment, the structure of the PGNN is as follows: Figure 4 As shown. Before failure, 20 points were uniformly selected from the magnetic flux density curve and stress-strain curve to form 20 sets of data (strain from 0 to 0.8). The input data were the normal magnetic flux density Mx and the tangential magnetic flux density My, and the output data was the bearing stress S. Since the tensile test was repeated 40 times, 800 sets of data were obtained, of which 680 sets were used for training and 120 sets were used for testing. The hidden layer consists of 10 neurons. To avoid weight conflicts in the forward path, the neurons in the hidden layer are only partially connected.

[0052] In one embodiment, the mathematical expression for a hidden layer neuron is:

[0053]

[0054]

[0055] The mathematical expression for the output layer neuron is:

[0056]

[0057]

[0058] in, This represents the weight of the j-th neuron in the L-th layer. This represents the bias of the j-th neuron in the L-th layer. This represents the linear result of the j-th neuron in the L-th layer. This represents the activation output of the j-th neuron in layer L.

[0059] The experimental results were analyzed and summarized into physical constraints, which will be discussed below. The physical constraints are summarized as follows: (1) As the normal magnetic induction intensity M y (2) As the tangential magnetic induction intensity M increases, the bearing stress S decreases; x As the load increases, the bearing stress S increases.

[0060] The physical constraints are expressed in terms of the derivatives of S with respect to My and Mx:

[0061]

[0062] Physical constraints are achieved through weights We add physical constraints to the neural network. To satisfy these physical constraints, we make each term in the physical constraint equation non-positive (non-negative) to ensure that S is relative to M. y and M x The derivatives of f1 and f2 are non-positive (non-negative). The derivatives of activation functions f1 and f2 are non-negative, therefore the physical constraints can be simply shown in Table 1.

[0063] Table 1. Correspondence between physical constraints

[0064]

[0065] For the current network and data, the mean squared error loss function exhibits good convergence during the debugging period. The mean squared error loss function is expressed as:

[0066]

[0067] Where S i It is predicted data, S it This is the test data, and n is the sample size.

[0068] In one embodiment, Figure 5 compares two types of composite material connection structures: a nano-coated smart composite fastener reinforced composite connection structure (NSCF connection structure) and an uncoated smart composite fastener reinforced composite connection structure (SCF connection structure).

[0069] All test samples of both NSCF and SCF connection structures underwent shear failure of the fasteners and bearing failure of the orifice. The difference lies in that the uncoated composite fasteners of the SCF connection structure experienced shear failure at the threads, while the nano-coated composite fasteners of the NSCF connection structure experienced shear failure at the screw head. As shown in Figure 5(c), processing defects such as fiber breakage and matrix crushing exist at the threads of the uncoated composite fasteners, which may be the initial damage points in the failure process.

[0070] Furthermore, the roughness of the thread surface caused by these defects also easily leads to stress concentration. Under load, the relative displacement of the upper and lower plates causes the composite fastener to tilt. Constrained by the nut and plate, the fastener bears shear loads, resulting in stress concentration at the thread defects. For nano-coated composite fasteners, the nano-coating improves the thread strength and makes the thread surface smoother, thereby reducing stress concentration. Therefore, nano-coated fasteners fracture at the screw head, while uncoated fasteners fracture at the thread.

[0071] As shown in Figure 5(a), the load-bearing strength of the NSCF connection structure is 196±12 MPa (repeated 40 times), while that of the SCF connection structure is 144±20 MPa (repeated 5 times). The higher average load-bearing strength of the NSCF connection structure is attributed to the improved load-bearing capacity and reduced stress concentration of the nano-coated threads. The larger error in the load-bearing strength of the SCF connection structure is due to the randomness of machining defects at the uncoated fastener threads. The higher load-bearing strength results in more severe load-bearing damage at the orifice, as shown in Figure 5(c). Furthermore, the stress-strain curves show that the stiffness of the NSCF connection structure is higher than that of the SCF connection structure, especially when the strain exceeds 0.5, as shown in Figure 5(a). This is because the stiffness of the connection structure is controlled by the stiffness of the plate and the fastener, and the nano-coated threads improve the stiffness of the fastener. When the strain exceeds 0.5, the decrease in stiffness of the SCF connection structure is due to the further expansion of damage to the uncoated fasteners.

[0072] As shown in Figure 5(b), the magnetic flux density of the NSCF connection structure before failure was recorded. The load-bearing stress was negatively correlated with the normal magnetic flux density My and positively correlated with the tangential magnetic flux density Mx. This phenomenon is due to the magneto-force effect and the tilting of the fasteners. The magnet embedded in the fasteners is subjected to shear loads during loading, thus the position of the magnetic domains tilts along the shear direction. In addition, the entire magnet tilts along with the tilting of the fasteners during loading. Therefore, the normal component My decreases and the tangential component Mx increases. This physical phenomenon serves as a constraint to guide the neural network, which is trained to predict the load-bearing stress of the connection structure and monitor the structural health.

[0073] In one embodiment, Figure 6(a) , 6(b) The PGNN was evaluated based on three aspects: prediction randomness, prediction accuracy, and training process. Six additional stretching experiments were conducted to generate test data. For comparison, a neural network (NN) was trained that was identical to the PGNN except for the absence of physical constraints.

[0074] Due to the initial random values ​​of the biases and weights before training, the load-bearing stress predicted by PGNN and NN is not constant. Both PGNN and NN were trained 20 times, with the same training data each time, but different initial random values ​​for the biases and weights. The results are shown in Figure 6(a). PGNN prediction results show good consistency. When using PGNN to predict load-bearing strength (maximum load-bearing stress), the maximum error of the 20 training results is 16.4 MPa. The average result of 20 repeated training iterations can reasonably predict the load-bearing stress. NN's prediction randomness is significant; all predicted stresses are smaller than the experimental results. When using NN to predict load-bearing strength, the maximum error of the 20 training results is 41.6 MPa, and the average of the 20 predicted load-bearing strengths is 37.5 MPa less than the experimental results. The proposed PGNN demonstrates the ability to reduce prediction randomness because physical constraints limit the direction of change of the initial random values.

[0075] The PGNN was trained 10 times using 680 sets of training data (34 tensile tests), and then its prediction accuracy was verified using 120 sets of test data (6 tensile tests). The results are shown in Figure 6(b). It can be seen that under the physical constraints of the PGNN, the results exhibit ideal characteristics, and the load-bearing stress predicted using magnetic induction intensity can reasonably quantify the true stress of the test sample. The maximum prediction error of the PGNN is 18.9 MPa. Figure 6(c) shows the loss values ​​of the PGNN and NN during training. For the PGNN, the loss value converged to below 1.5 within approximately 2000 cycles. However, for the NN, the loss value did not converge well until around 3500 cycles, indicating overfitting. This phenomenon demonstrates that physical constraints can accelerate the training process of neural networks and reduce the dependence on the amount of training data.

[0076] CF / PEEK composite fasteners and composite plates ([0 / +45 / -45 / 90]2s) short-cut carbon fibers according to Figure 7 The composite bolts and nuts are manufactured by pultrusion and machining, as shown in the method described.

[0077] In one embodiment, a nano-coating was designed to improve the hardness and wear resistance of the threads. For example... Figure 8As shown, the nanocoating consists of a solvent, a binder, and functional nanofillers. Polyimide (PI) is used as the solvent, dimethylformamide (DMF) as the binder, nano-graphite (8 μm) and nano-MoS2 (2 μm) are used to improve lubricity, nano-Al2O3 (20 μm) is used to improve hardness, and nano-CeF3 (3 μm) is used to improve wear resistance. The solid nanofillers are mixed with the completely dissolved binder and solvent and stirred for 30 minutes using a magnetic stirrer. The composite fasteners and nuts are immersed in the solution for 10 minutes and cured at 210°C for 50 minutes.

[0078] In order to monitor the health of the structure, such as Figure 2 As shown, a neodymium iron boron magnet is embedded in a nano-coated composite fastener. Based on the magneto-force effect, changes in magnetic flux density are recorded to predict structural stress. Normal and tangential magnetic flux densities are measured using a CH-3600 magnetic sensor.

[0079] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.

Claims

1. A smart composite fastener, characterized by, The threads of the fastener are coated with a nano-coating, and neodymium iron boron magnets are embedded in the shank of the fastener. Based on the magneto-force effect and a neural network containing physical constraints, a mapping between the magnetic induction intensity and the load-bearing stress of intelligent composite fasteners is established. By training a neural network with physical constraints, we can predict the load-bearing stress of the connected structure and monitor the structural health. Neural networks consist of an input layer, hidden layers, and an output layer. The training process includes: Multiple tensile tests were conducted on the composite material structure formed by the nano-coated composite fastener. Before failure, multiple data points are selected from the magnetic induction intensity curve and the stress-strain curve to form multiple sets of input data and output data, wherein the input data is the normal magnetic induction intensity M x and the tangential magnetic induction intensity M y , and the output data is the bearing stress S; The neural network is trained and tested using multiple sets of input and output data; The physical constraint equations are expressed in terms of the derivatives of the bearing stress S with respect to the tangential magnetic flux density My and the normal magnetic flux density Mx: ； By weight Weight constraints are added to incorporate into the neural network; each constraint is either non-positive or non-negative to ensure that the bearing stress S is relative to the tangential magnetic induction intensity M. y and normal magnetic induction intensity M x The derivatives of f1 and f2 are non-positive or non-negative; the derivatives of activation functions f1 and f2 are non-negative, and the weight constraints are: , j = 1 ~ 5; , j = 6 ~ 10.

2. The smart composite fastener of claim 1, wherein, The nano-coating includes polyimide, dimethylformamide, and functional nanofillers; the functional nanofillers include nano-graphite, nano-MoS2, nano-Al2O3, and nano-CeF3.

3. The smart composite fastener manufacturing process of claim 1, wherein, Dimethylformamide was completely dissolved in polyimide to form a solution. The functional nanofiller was stirred and mixed with the solution using a magnetic stirrer to form a mixture. The composite fasteners and nuts were then immersed in the mixture and cured at high temperature. After embedding neodymium iron boron magnets into the composite fasteners, they are sealed with sealant.

4. The smart composite fastener health monitoring method of claim 1, wherein, The mean squared error loss function of a neural network is expressed as: ； where S i is the predicted data, S it is the test data, and n is the sample size.

5. The intelligent composite material fastener health monitoring method according to claim 1, characterized in that, The mathematical expression for a hidden layer neuron is: ； ； The mathematical expression for the output layer neuron is: ； ； in, This represents the weight of the j-th neuron in the L-th layer. This represents the bias of the j-th neuron in the L-th layer. This represents the linear result of the j-th neuron in the L-th layer. This represents the activation output of the j-th neuron in layer L.

Images