Carbon fiber carbonization quality monitoring method based on resistance method
The carbon fiber carbonization quality monitoring method, which combines resistance method and deep learning model, solves the problem of online monitoring during carbon fiber carbonization, realizes stable monitoring of carbon fiber resistance value, and ensures high-quality production of composite materials.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- ZHEJIANG SCI-TECH UNIV
- Filing Date
- 2025-02-06
- Publication Date
- 2026-07-30
AI Technical Summary
Existing technologies cannot achieve online quality monitoring during the carbonization process of carbon fibers, resulting in excessive levels of hydrogen and oxygen residues, which affects the durability and performance of carbon fibers.
A resistance-based method for monitoring the carbonization quality of carbon fibers was adopted. By using a bridge circuit measuring instrument and conductive slip rings to wind carbon fiber bundles, combined with a deep learning model, the resistance value and quality of carbon fibers were monitored in real time by comparing the predicted resistance waveform with the measured resistance waveform.
This enables efficient and accurate quality monitoring of the carbon fiber carbonization process, ensuring the stability of the carbon fiber's resistivity and improving the quality and performance of the composite material.
Smart Images
Figure CN2025075907_30072026_PF_FP_ABST
Abstract
Description
A method for monitoring the carbonization quality of carbon fibers based on resistance method [Technical Field]
[0001] This invention relates to a method for monitoring the carbonization quality of carbon fibers, specifically a method for monitoring the carbonization quality of carbon fibers based on the resistance method, belonging to the field of quality detection technology in the carbon fiber production process. [Background Technology]
[0002] Carbon fiber, as an important component of high-performance composite materials, is widely used in aerospace, automotive industry, sporting goods, building materials and other fields due to its lightweight, high strength, high modulus and corrosion resistance.
[0003] The preparation process of carbon fiber typically includes pyrolysis and carbonization stages of the virgin fiber. During pyrolysis, the virgin fiber is exposed to a high-temperature environment, removing hydrogen and oxygen to form a high-carbon-content structure. The carbonization stage is a process carried out at even higher temperatures, further increasing the carbon content and crystallinity of the carbon fiber. Appropriate pyrolysis and carbonization temperatures can promote the removal of residual hydrogen and oxygen from the virgin fiber, while simultaneously increasing the fiber's carbon content and crystallinity. From the precursor fiber to after carbonization, the appearance and properties of the carbon fiber undergo significant changes, forming a carbon structure with high purity and crystallinity, possessing excellent mechanical and chemical properties.
[0004] However, insufficient carbonization of carbon fiber occurs when the residual hydrogen and oxygen in the precursor fiber exceeds the minimum standards for that stage of production. These issues can reduce its durability and performance. Therefore, to improve the quality of carbon fiber and ensure the manufacture of high-quality composite materials, strict quality monitoring is required during the carbon fiber production process. However, due to limitations in production equipment and the complexity of the operating environment, such testing must be conducted after carbon fiber production is completed, making online quality monitoring of the carbon fiber impossible.
[0005] Therefore, in order to solve the above-mentioned technical problems, it is indeed necessary to provide an innovative carbon fiber carbonization quality monitoring method based on resistance method to overcome the defects in the prior art. [Summary of the Invention]
[0006] The purpose of this invention is to provide a carbon fiber carbonization quality monitoring method based on resistance method, which can obtain the resistance value of carbon fiber with high efficiency and accuracy. The method monitors the quality of carbon fiber by comparing the waveform patterns of the trained model and the measured model obtained by deep learning with the resistance value of qualified carbon fiber samples.
[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a method for monitoring the carbonization quality of carbon fibers based on the resistance method, characterized by comprising the following steps:
[0008] 1) Prepare a bridge circuit measuring instrument and connect it to a pair of measuring axes via wiring;
[0009] 2) The carbon fiber bundles produced in the continuous production line in the carbonization furnace are wound in an S-shape onto the above-mentioned pair of measuring shafts;
[0010] 3) Start the bridge circuit measuring instrument. The bridge circuit measuring instrument automatically adjusts the value of its adjustable resistance to maintain the bridge balance. In the bridge balance state, the bridge circuit measuring instrument continuously measures the resistance value of each carbon fiber bundle between the two measuring axes.
[0011] 4) Since the resistance value of carbon fiber tow fluctuates during the production process, a deep learning framework is constructed to extract features related to the resistance value fluctuation and build a deep learning model. Then, two batches of several carbon fiber tows are randomly selected from the qualified samples that have been produced to measure the resistance value, which serves as the training set and validation set for the deep learning model, and the predicted resistance waveform is obtained. At the same time, based on the measured resistance value of the carbon fiber tow in step 3), the measured resistance waveform is obtained as the test set for the deep learning model.
[0012] 5) Compare the predicted resistance waveform from step 4) with the measured resistance waveform to determine if there are any quality problems with the carbon fiber bundle.
[0013] The carbon fiber carbonization quality monitoring method based on resistance method of the present invention is further described as follows: In step 1), the measuring shaft includes a shaft body, a conductive slip ring, and a brush; a bearing seat is provided on one side of the shaft body; the conductive slip ring is sleeved on the shaft body and is installed on the bearing seat by a number of slip ring fixing screws, so that the conductive slip ring can rotate relative to the shaft body; the carbon fiber bundle is wound on the conductive slip ring; one end of the brush abuts against the carbon fiber bundle, and the other end is electrically connected to the circuit.
[0014] The carbon fiber carbonization quality monitoring method based on the resistance method of the present invention is further described as follows: the conductive slip ring is provided with a plurality of partitions, and a thread is formed between adjacent partitions; a carbon fiber bundle passes through each thread and corresponds to a brush; a plurality of brushes are provided and are mounted on a brush fixing plate; each brush is provided with pressure by a spring so that it is always in contact with the carbon fiber bundle.
[0015] The carbon fiber carbonization quality monitoring method based on resistance method of the present invention further comprises: step 3) specifically: the resistance value of the carbon fiber bundle is measured using a four-wire system inside the bridge circuit measuring instrument, and the calculation formula is: R1 / R2=R3 / R x The resistance value of the carbon fiber bundle is the resistance R to be measured. x The rest are fixed resistors R1 and R2, and adjustable resistor R3;
[0016] The carbon fiber bundles move continuously in the carbonization furnace, that is, they move forward continuously over time during the processing. When the bridge becomes unbalanced, the bridge circuit measuring instrument automatically adjusts the value of the adjustable resistor R3 to maintain the bridge balance. In the balanced state of the bridge, the bridge circuit measuring instrument measures the resistance value of each carbon fiber bundle between the two continuously transmitted measuring axes.
[0017] The carbon fiber carbonization quality monitoring method based on the resistance method of the present invention is further described in step 4):
[0018] 4-1) Input the time series data of resistance values based on the simulation of the sine function, calculate the sine value of each point based on these time points; perform normalization preprocessing on the resistance value data to obtain the training set, validation set and test set;
[0019] 4-2) Define the network structure using a single-layer RNN and a fully connected layer; use MSELoss as the loss function in each training round of the RNN; update the model parameters through backpropagation, and train the neural network using gradient descent with the Adam optimizer. In each iteration, the model receives an input of one time step, performs forward and backward propagation, and updates the weight parameters.
[0020] 4-3) Use validation set data to evaluate model performance, adjust hyperparameters based on results, and retrain the model to optimize it; during the testing phase, the model uses the previously trained weights to recursively predict from the first data point of the sequence and outputs the final predicted resistance and measured resistance waveforms.
[0021] The carbon fiber carbonization quality monitoring method based on resistance method of the present invention further comprises: in step 4-1), a custom normalization process is performed on the measured resistance value data, and the calculation formula is: Where R nor This is the normalized resistance value, which is ultimately used as the amplitude of the sine wave. R is the actual resistance value. max and R min These are the maximum and minimum values of the resistance, respectively.
[0022] After obtaining the normalized resistance values, they are mapped onto a sine wave graph. The form of this sine wave is as follows: Where R nor f is the amplitude, f is the frequency, and t is the time. It is phase;
[0023] The sine value of each point is represented as: data = np.sin(time_steps), where sin() is a function in the NumPy library that calculates the sine value of each element in time_steps and stores the result in the variable data.
[0024] The carbon fiber carbonization quality monitoring method based on the resistance method of the present invention further comprises: in step 4-2), the MSELoss loss function is as follows:
[0025] In the formula, n is the number of samples, and y i It is the actual target value. It is the value predicted by the model; mean squared error is obtained by calculating the squared difference between the predicted value and the true value of each sample, and then averaging the squared errors of all samples to obtain an average error measure.
[0026] The carbon fiber carbonization quality monitoring method based on resistance method of the present invention further comprises: step 5) specifically: comparing the predicted resistance waveform based on the sinusoidal function simulation output in step 4) with the measured resistance waveform in terms of waveform shape, periodicity, and amplitude consistency; based on the maximum deviation error not exceeding 0.1%, if the predicted resistance waveform matches the actual performance of the measured resistance waveform and the deviation error is within 0.1%, it indicates that the carbonization quality of the carbon fiber bundle meets expectations and is in a qualified state; conversely, if the predicted resistance waveform deviates from the measured resistance waveform by more than 0.1%, there may be a potential problem of insufficient carbonization, which needs further investigation and verification.
[0027] The carbon fiber carbonization quality monitoring method based on the resistance method of the present invention further includes the following specific method for comparing the fluctuation patterns: First, check whether the shapes of the predicted waveform and the measured waveform are consistent. Since the waveform is simulated based on a sine function, the predicted resistance waveform should have a waveform shape that is extremely similar to that of the measured resistance waveform. Second, observe whether the time interval between the fluctuations of the two waveforms is consistent from the beginning of one cycle to the end of the next cycle. The predicted resistance waveform should be highly matched with the period of the measured resistance waveform. Finally, the amplitude and phase of the predicted resistance and measured resistance waveforms should be close. Check whether the maximum and minimum values of the amplitude are close and whether the waveform shows an upward or downward shift.
[0028] The carbon fiber carbonization quality monitoring method based on the resistance method of the present invention can also be described as follows: The deviation error is calculated as follows: For each corresponding time point within the same period, the difference between the predicted waveform resistance value and the measured waveform resistance value is calculated. By calculating the difference at each point, the entire waveform is quantitatively compared. From all the calculated relative errors, the maximum value is found, and the traditional detection method is used to verify whether the produced carbon fiber tow is qualified. The final maximum deviation error is 0.1%. The relative error is used in this method, and its formula is:
[0029] In the formula, R pre (t) is the predicted resistance value, R act (t) is the measured resistance value, based on which the relative error at each moment is calculated.
[0030] Compared with the prior art, the present invention has the following beneficial effects: The carbon fiber carbonization quality monitoring method based on the resistance method of the present invention combines the measurement of resistance by a bridge circuit and the winding of carbon fiber bundles by a conductive slip ring. It can efficiently and accurately monitor the resistance value during the carbon fiber production process. By comparing the predicted resistance waveform with the measured resistance waveform based on the deep learning network structure, the quality of carbon fiber during the production process can be monitored, so as to manufacture high-quality composite materials. [Attached Image Description]
[0031] Figure 1 is a front view of the carbon fiber production line of the present invention.
[0032] Figure 2 is a schematic diagram of the device for monitoring the carbon fiber carbonization quality based on the resistance method of the present invention.
[0033] Figure 3 is a top view of the measurement axis in Figure 2.
[0034] Figure 4 is a waveform diagram showing that the predicted resistance and the measured resistance match in the method of the present invention.
[0035] Figure 5 is a waveform diagram showing the deviation between the predicted resistance and the measured resistance in the method of the present invention.
[0036] Figure 6 is a waveform diagram showing the large fluctuations between the predicted resistance and the measured resistance when using the method of the present invention.
[0037] Figure 7 is a flowchart of generating the waveforms of the predicted resistance and the measured resistance using the method of the present invention.
Detailed Implementation Methods
[0038] Please refer to Figures 1 to 7 in the specification. This invention is a method for monitoring the carbonization quality of carbon fibers based on the resistance method, which includes the following steps:
[0039] 1) Prepare a bridge circuit measuring instrument 10 and connect the bridge circuit measuring instrument 10 to a pair of measuring axes 1 via a circuit.
[0040] The bridge circuit measuring instrument 10 has multiple measurement ports for simultaneous detection of multiple carbon fiber bundles 2. In this embodiment, the bridge circuit measuring instrument 10 can simultaneously monitor the resistance of six carbon fiber bundles 2. The positive and negative terminals of the power supply in the bridge circuit measuring instrument 10 must be correctly connected to the bridge body to ensure power supply and overall operation of the bridge. The galvanometer in the bridge circuit measuring instrument 10 is used to measure minute current changes when the bridge is balanced. Its two leads need to be connected to the two galvanometer ports of the bridge body respectively to detect and record the current changes of the bridge in a timely manner.
[0041] 2) The carbon fiber bundle 2 produced in the carbonization furnace 5 is wound in an S-shape onto the pair of measuring shafts 1.
[0042] The carbonization furnace 5 is equipped with several pairs of ordinary rotating shafts 3 and the measuring shaft 1, and the measuring shaft 1 is located above the ordinary rotating shafts 3, through which the carbon fiber bundle 2 is continuously conveyed.
[0043] Furthermore, the measuring shaft 1 is composed of several parts, including a shaft body 12, a conductive slip ring 9, and a brush 6. A bearing seat 8 is provided on one side of the shaft body 12; the conductive slip ring 9 is sleeved on the shaft body 12 and is mounted on the bearing seat 8 by several slip ring fixing screws 7, allowing the conductive slip ring 9 to rotate relative to the shaft body 1. The carbon fiber bundle 2 is wound around the conductive slip ring 9. One end of the brush 6 abuts against the carbon fiber bundle 2, and the other end is electrically connected to the circuit, forming a circuit between the carbon fiber bundle 2, the measuring shaft 1, and the bridge circuit measuring instrument 10.
[0044] The conductive slip ring 9 is provided with several partitions 13, and a thread channel 14 is formed between adjacent partitions 13; a carbon fiber bundle 2 passes through each thread channel 14 and corresponds to a brush 6. The partitions 13 can prevent mutual interference between the carbon fiber bundles 2.
[0045] Several brushes 6 are provided and mounted on a brush fixing plate 11. Each brush 6 is pressurized by a spring to ensure it is always in electrical contact with the carbon fiber bundle 2 and to maintain a continuous connection between the brush 6 and the carbon fiber bundle 2 during production. For clarity of the wiring in Figure 2, the circuit connects only one pair of brushes 6 to the bridge circuit measuring instrument 10. Furthermore, the spring force can be adjusted to adapt to different friction conditions and wear situations.
[0046] The design principle of the measuring shaft 1 is as follows: the carbon fiber bundle 2 is wound around the conductive slip ring 9, and the continuous feeding of the carbon fiber bundle 2 is achieved by the rotation of the conductive slip ring 9. The brush 6 abuts against the carbon fiber bundle 2 and supplies power to the carbon fiber bundle 2 located between a pair of measuring shafts 1, so that the bridge circuit measuring instrument 10 can measure the resistance of the carbon fiber bundle 2.
[0047] 3) Start the bridge circuit measuring instrument 10. The bridge circuit measuring instrument 10 automatically adjusts the value of its adjustable resistor to maintain bridge balance. In the bridge balanced state, the bridge circuit measuring instrument 10 continuously measures the resistance value of each carbon fiber bundle 2 between the two measuring axes 1.
[0048] Specifically, the bridge circuit measuring instrument 10 uses a four-wire system to measure the resistance of the carbon fiber bundle 2, and the calculation formula is: R1 / R2=R3 / R x The resistance value of the carbon fiber bundle is the resistance R to be measured. x The rest are fixed resistors R1 and R2, and adjustable resistor R3.
[0049] The carbon fiber tow 2 operates continuously in the carbonization furnace 5, meaning it moves forward continuously over time during processing. When the bridge becomes unbalanced, the bridge circuit measuring instrument 10 automatically adjusts the value of the adjustable resistor R3 to maintain bridge balance. In a balanced state, the bridge circuit measuring instrument 10 measures the resistance value of each carbon fiber tow 2 between the two continuously moving measuring axes.
[0050] 4) Since the resistance value of carbon fiber tow 2 fluctuates due to the rotation and conveying of the ordinary rotating shaft 3 during the production process, features related to resistance fluctuation are extracted by constructing a deep learning flowchart framework as shown in Figure 7. Two batches of carbon fiber tows with a length of 1 meter are randomly selected from the produced qualified samples to measure their resistance values, which serve as the input data for the deep learning model, i.e., the training set and validation set. Through training, the predicted resistance waveform based on sine function simulation, as shown by the dashed lines in Figures 4 to 6, is obtained. Simultaneously, based on the resistance value of carbon fiber tow 2 measured by the instrument in step 3), as the test set for the deep learning model, the measured resistance waveform based on sine function simulation, as shown by the solid lines in Figures 4 to 6, is constructed. The specific process is as follows:
[0051] 4-1) Input the time series data of resistance values based on the simulation of the sine function, calculate the sine value of each point based on these time points; perform normalization preprocessing on the resistance value data to obtain the training set, validation set and test set.
[0052] The measured resistance values are normalized using the following formula: R norThis is the normalized resistance value, which is ultimately used as the amplitude of the sine wave. R is the actual resistance value. max and R min These are the maximum and minimum values of the resistance, respectively.
[0053] After obtaining the normalized resistance values, they are mapped onto a sine wave graph. The form of this sine wave is as follows: Where R nor As the amplitude of the waveform, f is the frequency, and t is the time. It is phase;
[0054] The sine value for each point is represented as: data = np.sin(time_steps). Based on this, the final output is a waveform graph of the predicted resistance and the measured resistance simulated based on the sine function. sin() is a function in the NumPy library that calculates the sine value of each element in time_steps and stores the result in the variable data.
[0055] 4-2) Define the network structure using a single-layer RNN and a fully connected layer. In each training iteration of the RNN, use MSELoss (mean squared error loss function) as the loss function. Update the model parameters through backpropagation and train the neural network using gradient descent with the Adam optimizer. In each iteration, the model receives an input of one time step, performs forward and backward propagation, and updates the weight parameters.
[0056] The MSELoss loss function is as follows:
[0057] In the formula, is the number of samples, is the true target value, and is the value predicted by the model. The mean squared error is calculated by averaging the squared errors of all samples, resulting in an average error measure.
[0058] The Adam optimizer maintains two momentum values for each parameter, as follows:
[0059] First-order moment estimate: m t =β1m t-1 +(1-β1)g t ;
[0060] In the formula, m t It is the first moment estimate of the gradient at the current time, g t β1 is the current gradient, β1 is the decay rate of the first moment, and m is the current gradient. t-1 It is the estimated value from the previous moment.
[0061] Second moment estimate: v t =β2vt-1 +(1-β2)g t 2 ;
[0062] In the formula, v t This is the gradient second moment estimate at the current moment, where β2 is the decay rate of the second moment, and v t-1 It is the estimated value from the previous moment.
[0063] Deviation correction:
[0064] Parameter update:
[0065] In the formula, θ t These are the model parameters at the current time step, α is the learning rate, and ε is a small constant added to prevent the denominator from being zero.
[0066] 4-3) Use validation set data to evaluate model performance, adjust hyperparameters (such as learning rate, hidden layer size, etc.) based on the results, and retrain the model to optimize it; during the testing phase, the model uses the previously trained weights to recursively predict from the first data point of the sequence and outputs the final predicted resistance and measured resistance waveforms.
[0067] 5) Compare and verify the fluctuation pattern of the predicted resistance waveform in step 4) with the measured resistance waveform to determine whether there is a quality problem with the carbon fiber bundle 2.
[0068] Specifically, the predicted resistance waveform output based on the sine function simulation in step 4) is compared with the measured resistance waveform in terms of waveform shape, periodicity, and amplitude consistency. First, check if the shapes of the predicted and measured waveforms are consistent. Since the waveform is simulated based on a sine function, the predicted resistance waveform should have a very similar shape to the measured resistance waveform. Second, observe whether the time interval between the fluctuations of the two waveforms is consistent from the beginning of one cycle to the end of the next cycle. The predicted resistance waveform should have a high degree of periodicity matching with the measured resistance waveform. Finally, the amplitude and phase of the predicted and measured resistance waveforms should be close. Check if the maximum and minimum amplitude values are close and whether the waveform shows any upward or downward shift.
[0069] After calculation and verification of the deviation error between the predicted resistance waveform and the measured resistance waveform, it was determined that the maximum deviation error should not exceed 0.1%. The specific calculation method for the deviation error is as follows: For each corresponding time point within the same period, the difference between the predicted waveform resistance value and the measured waveform resistance value is calculated. By calculating the difference at each point, the entire waveform is quantitatively compared. From all calculated relative errors, the maximum value is found, and traditional detection methods are used to verify whether the produced carbon fiber tow is qualified. The final maximum deviation error is 0.1%. This method uses relative error calculation, and its formula is:
[0070] In the formula, R pre (t) is the predicted resistance value, R act (t) is the measured resistance value, based on which the relative error at each moment is calculated.
[0071] As shown in Figure 4, if the predicted resistance waveform and the measured resistance waveform are basically consistent and the error is within 0.1%, it indicates that the carbonization quality of the carbon fiber bundle 2 meets expectations and is in a qualified state. Conversely, as shown in Figures 5 and 6, if the measured resistance waveform exhibits waveform distortion, waveform shift, or local waveform distortion, and there is a deviation of more than 0.1% compared with the predicted resistance waveform, there may be a potential problem of insufficient carbonization, which requires further investigation and verification.
[0072] In summary, the carbon fiber carbonization quality monitoring method based on the resistance method of the present invention combines the measurement of resistance by a bridge circuit and the winding of carbon fiber bundles by a conductive slip ring. It can efficiently and accurately monitor the resistance value during the carbon fiber production process. By comparing the predicted resistance waveform with the measured resistance waveform based on a deep learning model, the key parameters of the carbon fiber bundle during the carbonization process can be monitored in real time, providing an effective assessment of process stability and product quality.
[0073] The above-described specific embodiments are merely preferred embodiments of this invention and are not intended to limit this invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the protection scope of this invention.
Claims
1. A method for monitoring the carbonization quality of carbon fibers based on resistance method, characterized in that: Includes the following steps: 1) Prepare a bridge circuit measuring instrument and connect it to a pair of measuring axes via wiring; 2) The carbon fiber bundles produced in the continuous production line in the carbonization furnace are wound in an S-shape onto the above-mentioned pair of measuring shafts; 3) Start the bridge circuit measuring instrument. The bridge circuit measuring instrument automatically adjusts the value of its adjustable resistance to maintain the bridge balance. In the bridge balance state, the bridge circuit measuring instrument continuously measures the resistance value of each carbon fiber bundle between the two measuring axes. 4) Since the resistance value of carbon fiber tow fluctuates during the production process, a deep learning framework is constructed to extract features related to the resistance value fluctuation and build a deep learning model. Then, two batches of several carbon fiber tows are randomly selected from the qualified samples that have been produced to measure the resistance value, which serves as the training set and validation set for the deep learning model, and the predicted resistance waveform is obtained. At the same time, based on the measured resistance value of the carbon fiber tow in step 3), the measured resistance waveform is obtained as the test set for the deep learning model. 5) Compare the predicted resistance waveform from step 4) with the measured resistance waveform to determine if there are any quality problems with the carbon fiber bundle.
2. The carbon fiber carbonization quality monitoring method based on resistance method as described in claim 1, characterized in that: In step 1), the measuring shaft includes a shaft body, a conductive slip ring, and a brush; a bearing seat is provided on one side of the shaft body; the conductive slip ring is sleeved on the shaft body and is installed on the bearing seat by several slip ring fixing screws, so that the conductive slip ring can rotate relative to the shaft body; the carbon fiber bundle is wound on the conductive slip ring; one end of the brush abuts against the carbon fiber bundle, and the other end is electrically connected to the circuit.
3. The carbon fiber carbonization quality monitoring method based on the resistance method as described in claim 2, characterized in that: The conductive slip ring is provided with several partitions, and a thread is formed between adjacent partitions; a carbon fiber bundle passes through each thread and corresponds to a brush; several brushes are provided and are mounted on a brush fixing plate; each brush is pressured by a spring so that it is always in contact with the carbon fiber bundle.
4. The carbon fiber carbonization quality monitoring method based on resistance method as described in claim 1, characterized in that: Step 3) specifically involves: the internal four-wire system of the bridge circuit measuring instrument is used to measure the resistance of the carbon fiber bundle, and the calculation formula is: R1 / R2=R3 / R x , The resistance value of the carbon fiber bundle is the resistance R to be measured. x The rest are fixed resistors R1 and R2, and adjustable resistor R3; The carbon fiber bundles move continuously in the carbonization furnace, that is, they move forward continuously over time during the processing. When the bridge becomes unbalanced, the bridge circuit measuring instrument automatically adjusts the value of the adjustable resistor R3 to maintain the bridge balance. In the balanced state of the bridge, the bridge circuit measuring instrument measures the resistance value of each carbon fiber bundle between the two continuously transmitted measuring axes.
5. The carbon fiber carbonization quality monitoring method based on resistance method as described in claim 1, characterized in that: Step 4) specifically involves: 4-1) Input the time series data of resistance values based on the simulation of the sine function, calculate the sine value of each point based on these time points; perform normalization preprocessing on the resistance value data to obtain the training set, validation set and test set; 4-2) Define the network structure using a single-layer RNN and a fully connected layer; use MSELoss as the loss function in each training round of the RNN; update the model parameters through backpropagation, and train the neural network using gradient descent with the Adam optimizer. In each iteration, the model receives an input of one time step, performs forward and backward propagation, and updates the weight parameters. 4-3) Use validation set data to evaluate model performance, adjust hyperparameters based on results, and retrain the model to optimize it; During the testing phase, the model uses the previously trained weights to recursively predict from the first data point in the sequence and outputs the final predicted resistance and measured resistance waveforms.
6. The carbon fiber carbonization quality monitoring method based on resistance method as described in claim 5, characterized in that: Step 4-1) specifically involves: performing a custom normalization process on the measured resistance data. The calculation formula is as follows: Where R nor This is the normalized resistance value, which is ultimately used as the amplitude of the sine wave. R is the actual resistance value. max and R min These are the maximum and minimum values of the resistance, respectively. After obtaining the normalized resistance values, they are mapped onto a sine wave graph; the form of this sine wave is: Where R nor f is the amplitude, f is the frequency, and t is the time. It is phase; The sine value of each point is represented as: data = np.sin(time_steps), where sin() is a function in the NumPy library that calculates the sine value of each element in time_steps and stores the result in the variable data.
7. The carbon fiber carbonization quality monitoring method based on the resistance method as described in claim 5, characterized in that: In step 4-2), the MSELoss loss function is as follows: In the formula, n is the number of samples, and y i It is the actual target value. It is the value predicted by the model; mean squared error is obtained by calculating the squared difference between the predicted value and the true value of each sample, and then averaging the squared errors of all samples to obtain an average error measure.
8. The carbon fiber carbonization quality monitoring method based on resistance method as described in claim 1, characterized in that: Step 5) specifically involves comparing the predicted resistance waveform based on the sinusoidal function simulation output in step 4) with the measured resistance waveform in terms of waveform shape, periodicity, and amplitude consistency to determine the fluctuation pattern. Based on the premise that the maximum deviation error should not exceed 0.1%, if the predicted resistance waveform and the measured resistance waveform match the actual performance and the deviation error is within 0.1%, it indicates that the carbonization quality of the carbon fiber bundle meets expectations and is in a qualified state; conversely, if the predicted resistance waveform and the measured resistance waveform deviate by more than 0.1%, there may be a potential problem of insufficient carbonization, which requires further investigation and verification.
9. The carbon fiber carbonization quality monitoring method based on resistance method as described in claim 8, characterized in that: The specific method for comparing the fluctuation patterns is as follows: First, check whether the shapes of the predicted waveform and the measured waveform are consistent. Since the waveform is simulated based on a sine function, the predicted resistance waveform should have a very similar waveform shape to the measured resistance waveform. Second, observe whether the time interval between the fluctuations of the two waveforms from the beginning of one cycle to the end of the next cycle is consistent. The period of the predicted resistance waveform should be highly matched with that of the measured resistance waveform. Finally, the amplitude and phase of the predicted resistance and measured resistance waveforms should be close. Check whether the maximum and minimum values of the amplitude are close and whether the waveform shows an upward or downward shift.
10. The carbon fiber carbonization quality monitoring method based on resistance method as described in claim 8, characterized in that: The deviation error is calculated as follows: For each corresponding time point within the same period, the difference between the predicted waveform resistance value and the measured waveform resistance value is calculated. By calculating the difference at each point, the entire waveform is quantitatively compared. From all the calculated relative errors, the maximum value is found, and the production of carbon fiber tow is verified using traditional detection methods. The final maximum deviation error is 0.1%. The method uses relative error for calculation, and its formula is: In the formula, R pre (t) is the predicted resistance value, R act (t) is the measured resistance value, based on which the relative error at each moment is calculated.