Electrolyte concentration monitoring device and electrolyte concentration monitoring system
Through optical detection and machine learning algorithm combined with spectral analysis technology, the problem of rapid measurement of electrolyte concentration is solved, efficient and accurate monitoring of electrolyte concentration is achieved, and the accuracy and stability of the electroplating process are improved.
Patent Information
- Application Number
- CN202510504565.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-22
- Publication Date
- 2025-07-11
AI Technical Summary
In the prior art, it is difficult to measure the electrolyte concentration quickly, and the electrolyte replenishment rate cannot be accurately controlled, resulting in a decrease in electroplating accuracy.
The optical signal emitter and total reflector combined with spectral analysis technology are used to calculate the electrolyte concentration through optical detection means, and the spectral information is accurately converted into particle concentration using machine learning algorithms and support vector machine models. The spectral characteristics are extracted in combination with the convolutional network model to achieve efficient and accurate electrolyte concentration monitoring.
Real-time monitoring and dynamic control of electrolyte concentration is realized, the accuracy and efficiency of electroplating are improved, and the uniformity and stability of the coating quality are ensured.
Smart Images
Figure CN120293914A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of electrolytes, and in particular, to an electrolyte concentration monitoring device and an electrolyte concentration monitoring system. Background Art
[0002] Electroplating is a process of depositing a metal layer on the surface of an object using the principle of electrolysis. Its core is to drive metal ions to be reduced on the surface of the substrate through an electric current to form a uniform and dense coating. The concentration of the electrolyte directly affects the coating quality and process stability. When the electrolyte concentration is too low and the number of ions is insufficient, the cathodic reaction rate will decrease, the ion migration speed will be limited, uneven deposition is likely to occur, and even "burning" (locally too thick and loose structure) will appear, and the effective reaction will decrease, and electrical energy will be wasted on side reactions (such as hydrogen evolution). When the electrolyte concentration is too high, the ion reduction will be accelerated, resulting in too fast formation of crystal nuclei, forming a rough surface, and ion accumulation near the anode and ion depletion near the cathode will occur, leading to uneven coating thickness.
[0003] Therefore, it is necessary to control the concentration of the electrolyte in the electroplating process. In the prior art, generally, electrolytes are supplemented according to the reaction rate, so as to ensure that the concentration of the electrolyte is within a reasonable range. In practice, such a scheme requires accurate measurement of the electrolyte concentration. However, it is difficult to quickly measure the electrolyte concentration. Only through sampling and then using chemical titration experiments, the electrolyte concentration cannot be quickly measured, and thus the replenishment rate of the electrolyte cannot be accurately controlled, resulting in a reduction in the electroplating accuracy. Summary of the Invention
[0004] The content part of the present application is used to briefly introduce concepts, which will be described in detail in the following specific implementation part. The content part of the present application is not intended to identify the key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.
[0005] In order to solve the technical problems mentioned in the above background art part, some embodiments of the present application provide an electrolyte concentration monitoring device, including:
[0006] An optical signal transmitter for emitting optical signals of different wavelengths;
[0007] A sampling tube provided with a sampling cavity for placing the electrolyte;
[0008] A total reflector disposed at the bottom of the sampling tube;
[0009] An optical signal decoder for receiving the optical signal and analyzing the wavelength of the optical signal;
[0010] Among them, the optical signal transmitter is arranged at the end of the sampling tube and emits detection light rays into the sampling chamber. The detection light rays pass through the electrolyte and irradiate onto the total reflector to form a reflection signal. The optical signal decoder receives the reflection signal and calculates the concentration of corresponding particles in the electrolyte according to the wavelength shift of the reflection signal.
[0011] In the technical solution provided by this application, according to the influence relationship between the change of particle concentration in the electrolyte on the refractive index, the electrolyte concentration is calculated by optical detection means, which has the characteristics of high detection efficiency and high precision, and can monitor the dynamic change of the electrolyte concentration in real time.
[0012] The emission light and the reflected light share the same optical path, which will affect the layout positions of the optical signal decoder and the optical signal transmitter. It is easy to cause the optical path to deflect, thereby affecting the reception of the optical signal. For this reason, this application provides the following technical solution:
[0013] A circulator is arranged at the end of the sampling tube, the common port of the circulator faces the bottom of the sampling tube, and the first inlet and the second inlet of the circulator face the optical signal decoder and the total reflector respectively.
[0014] In the technical solution provided by this application, the circulator is used for beam splitting operation. Thus, when the reflected light passes through the circulator, it will not enter the optical signal transmitter along the original optical path, but will be emitted from another inlet to the optical signal decoder. In this way, the positions of the optical signal decoder and the optical signal transmitter can be independent of each other, avoiding distortion of the received optical signal.
[0015] The spectral information cannot accurately reflect the particle concentration. When converting the change situation of the spectral information into the particle concentration, it cannot be accurately converted according to the information relationship between the two. For this reason, this application provides the following technical solution:
[0016] The optical signal decoder includes:
[0017] An optical signal receiving unit for receiving optical signals;
[0018] A spectrum reading unit for reading the received optical signal to obtain spectral information;
[0019] A particle concentration calculation unit for converting the spectral information into the particle concentration based on a machine learning algorithm.
[0020] In the technical solution provided by this application, by using the machine learning algorithm, the internal relationship between the spectral information and the particle concentration in the data sample can be obtained, so as to find the law between the particle concentration and the spectral information. In this way, the particle concentration can be accurately measured.
[0021] Although spectral information can accurately reflect the refractive index of a solution, the refractive index is affected by various factors. Directly using spectral information to calculate the particle concentration is not highly accurate. Therefore, the present application provides the following technical solutions:
[0022] Furthermore, the optical signal decoder further includes an information acquisition unit; the information acquisition unit is used to acquire the temperature of the electrolyte and the particle composition of the electrolyte, and make the temperature of the electrolyte and the particle composition of the electrolyte into external information, and send it to the particle concentration calculation unit.
[0023] In the technical solutions provided by the present application, making the temperature of the electrolyte and the particle composition of the electrolyte into external information can lead to higher accuracy when converting spectral information into particle concentration.
[0024] Spectral information actually cannot directly reflect the particle concentration. Especially when the particle types in the electrolyte are complex, directly using a machine learning model to learn the relationship between spectral information and particle concentration will cause the model to overfit the data, and thus the accuracy is not high in actual particle concentration measurement. Therefore, the present application provides the following technical solutions:
[0025] Furthermore, the particle concentration calculation unit is built-in with a convolutional network model and a support vector machine model;
[0026] Among them, the convolutional network model is used to convert spectral information into refractive index;
[0027] The support vector machine model is used to convert the refractive index and external information into particle concentration.
[0028] In the technical solutions provided by the present application, the particle concentration calculation unit is built-in with two models. When the convolutional network model processes spectral information, it can quickly extract the characteristic changes in the spectral information, and then accurately convert the spectral information into refractive index. When calculating the particle concentration, the support vector machine model is used. The support vector machine model can fuse multi-dimensional information and accurately find the corresponding particle concentration range through classification, so as to accurately extract the particle concentration.
[0029] The principle that spectral information reflects the refractive index is that the change in the refractive index will cause the light intensity of the optical signal to change. When optical signals of different frequencies are refracted, they are affected differently, so the light intensity will also change. However, relying solely on this information cannot intuitively reflect the relationship between the refractive index and the optical signal.
[0030] Therefore, the present application provides the following technical solutions:
[0031] Furthermore, taking the wavelength in the spectral information as time-domain information, the spectral information is converted into spectral density information.
[0032] In the technical solution provided by this application, spectral information is converted into frequency spectrum information, in which the intensity change under different frequency components can be more intuitively reflected, so the correlation between the refractive index and the optical signal can be better reflected.
[0033] To judge the particle concentration in the electrolyte according to the refractive index of the electrolyte, it is necessary to accurately calculate the refractive index of the electrolyte to increase the information content of the refractive index. However, in the spectral information, the change information generated by the refractive index in a small range has very little effect on the light intensity. In practice, the resolution accuracy is not high. Therefore, this application provides the following technical solution:
[0034] Further, the following steps are used to convert the spectral information into frequency spectrum information;
[0035] S1: Set the initial sampling rate f s , and convert the original spectral information into a discrete signal x[n], where n is the index of the sampling point;
[0036] S2: Divide the discrete signal x[n] into m signal frames;
[0037] x m [n] = x[n + m*M], where N is the length of the signal frame, M is the overlapping degree of the signal frames, 0 < n < N, and m represents the index of the interval;
[0038] S3: Perform a discrete Fourier transform on each signal frame to obtain the frequency spectrum X m [k];
[0039]
[0040] Among them, k represents the frequency index, and e represents the natural constant;
[0041] S4: Calculate the power spectrum P m [k];
[0042] P m [k] = |X m [k]| 2 = X m [k] × X m * [k], X m * [k] is the complex conjugate of X m [k];
[0043] S5: Calculate the power spectrum: log P m [k];
[0044] log P m [k] = log(|Xm [k]| 2 +∈);
[0045] where ε represents a minimum value;
[0046] S6: Perform an inverse Fourier transform on the power spectrum to obtain cepstrum coefficients c m [n], arrange the cepstrum coefficients in a matrix according to time to generate spectral information;
[0047] where j is the imaginary unit, j 2 = -1, and k is the frequency index.
[0048] In the technical solution provided by this application, after converting spectral information into discrete information, through cepstrum operations, the fundamental frequency and resonance peak information in the spectral information can be extracted. Furthermore, the sensitivity to amplitude changes at different wavelengths in the spectral information can be further enhanced, thereby increasing the ability to distinguish spectral information and further increasing the resolution ability of the refractive index.
[0049] When the convolutional layer extracts information, it will inevitably lose some information. If the lost information is forcibly increased, it will cause the convolutional layer to be unable to remove redundant information, resulting in an excessive consumption of computing resources by the convolutional layer. To address this problem, this application provides the following technical solution:
[0050] Furthermore, the convolutional network model includes:
[0051] An input layer for inputting spectral information;
[0052] A convolutional layer including 3 parallel convolutional blocks, and the number of filters in the convolutional blocks gradually increases;
[0053] A fully connected layer for receiving the outputs of each convolutional layer to obtain combined features;
[0054] An output layer for classifying the combined features into refractive index labels.
[0055] In the technical solution provided by this application, the number of filters in each convolutional layer is different, and thus different features can be captured. The convolutional block with a small number of filters can capture the change information in the short frequency of the spectral information and capture the intensity displacement in the spectral information in a short time, while the convolutional block with a large number of filters can extract high-level abstract features and find the intensity offset law in the spectral information from a global perspective.
[0056] Furthermore, the support vector machine model calculates the particle concentration using the following steps;
[0057] Z1: Obtain the particles whose concentrations in the electrolyte will change based on external information, and generate concentration labels for these particles; construct the original problem according to the number of concentration labels;
[0058] y i (wx i +b) > 1, or y i (wx i +b) < -1; i = 1, 2... n, n is an integer, n > 1, i
[0059] represents the sample index, w is the normal vector of the hyperplane, b is the intercept of the hyperplane, x i is the feature vector of the i-th sample, y i is the set of corresponding class labels;
[0060] Z2: Introduce the Lagrangian function:
[0061] i and j represent the sample indices, α, α i 、α j respectively represent the Lagrange multipliers, L represents the Lagrangian function, and K represents the kernel function;
[0062] Z3: Solve the dual problem to find the optimal Lagrange multiplier α i ;
[0063]
[0064] max a w(α) represents the value of the objective function w(α) maximized with respect to the Lagrange multiplier α, and w(α) is the objective function of the dual problem corresponding to the original optimization problem;
[0065] Subject to the following constraints:
[0066]
[0067] 0 ≤ α i ≤ C, i = 1...... n,
[0068] K(x i , x j ) is the kernel function used to calculate the inner product between samples, and C represents the penalty function;
[0069] Z4: Construct a decision function for prediction based on the optimal α i :
[0070] where sign is the label and x represents the unknown sample to be classified.
[0071] In the technical solution provided by this application, a support vector machine is used to calculate the particle concentration in the electrolyte, and the corresponding concentration label can be set according to the required measurement accuracy, so that the characteristic information can be regressed to the corresponding concentration label while meeting the production requirements, increasing the efficiency of concentration calculation.
[0072] The present invention also provides an electrolyte concentration monitoring system, including an information processor and a plurality of electrolyte concentration monitoring devices; the plurality of electrolyte concentration monitoring devices are evenly arranged at various positions in the electrolyte, and the information processor uses the average concentration of the particle concentrations measured at each position in the electrolyte as the concentration information of the particles in the electrolyte. BRIEF DESCRIPTION OF THE DRAWINGS
[0073] The drawings forming a part of this application are used to provide a further understanding of this application, making other features, objectives, and advantages of this application more obvious. The schematic embodiments and their descriptions of this application are used to explain this application and do not constitute an improper limitation of this application.
[0074] In addition, throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic, and the elements and elements are not necessarily drawn to scale.
[0075] In the drawings:
[0076] Figure 1 is a schematic structural diagram of the electrolyte concentration monitoring device.
[0077] Figure 2 is a wavelength-light intensity change diagram. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0078] The embodiments of this application will be described in more detail below with reference to the drawings. Although some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments described herein. On the contrary, these embodiments are provided to more thoroughly and completely understand this application. It should be understood that the drawings and embodiments of this application are only for exemplary purposes and are not used to limit the protection scope of this application.
[0079] In addition, it should be noted that only the parts related to the relevant invention are shown in the drawings for the convenience of description. Without conflict, the embodiments in this application and the features in the embodiments can be combined with each other.
[0080] This application will be described in detail below with reference to the drawings and in combination with the embodiments.
[0081] Refer to Figure 1The electrolyte concentration monitoring device includes an optical signal transmitter, a sampling tube, a total reflector, and an optical signal decoder. The optical signal transmitter is used to transmit optical signals of different wavelengths. The sampling tube is provided with a sampling cavity for placing the electrolyte. The total reflector is arranged at the bottom of the sampling tube. The optical signal decoder is used to receive the optical signal and analyze the wavelength of the optical signal. The optical signal transmitter is arranged at the end of the sampling tube, and emits a detection light to the sampling cavity. The detection light passes through the electrolyte and irradiates the total reflector to form a reflection signal. The optical signal decoder receives the reflection signal and calculates the corresponding particle concentration in the electrolyte according to the wavelength offset of the reflection signal.
[0082] The optical signal transmitter uses a multi-wavelength tunable laser light source with a wavelength range of 400nm to 1600nm and an output power of 5mW±0.1mW. The light source module has a built-in temperature control device to ensure that the wavelength stability error is less than ±0.2nm. The optical signal is transmitted to the end of the sampling tube through a single-mode optical fiber, and the anti-reflection coating on the end face of the optical fiber can reduce reflection loss.
[0083] The sampling tube is a cylindrical quartz glass tube with an inner diameter of 6mm and a length of 60mm. Both ends are sealed with optical-grade epoxy resin. The inside of the sampling cavity is polished to a surface roughness of Ra≤0.1μm to reduce light scattering. The total reflector is installed at the bottom of the sampling tube and is composed of a high-refractive-index magnesium fluoride (MgF2) prism. The incident angle is designed to be 68° (greater than the critical angle of 65°) to ensure total reflection conditions. A notch is opened on the side wall of the sampling tube, and the sampling tube is placed in the electrolyte. The electrolyte enters the sampling tube through the notch, so that after the optical signal enters the sampling tube, it needs to be scattered by the electrolyte before it can return to the optical signal decoder.
[0084] The transmitted light and the reflected light share the same optical path, which will affect the layout of the optical signal decoder and the optical signal transmitter. It is easy to cause the optical path to deflect, thereby affecting the reception of the optical signal. To this end, the present application provides the following technical solutions:
[0085] The end of the sampling tube is provided with a circulator, the common port of the circulator faces the bottom of the sampling tube, and the first inlet and the second inlet of the circulator face the optical signal decoder and the total reflector respectively. The first inlet is connected to the optical signal transmitter through an optical fiber. The second inlet is connected to the optical signal decoder through an optical fiber.
[0086] The circulator uses a three-port fiber circulator with an insertion loss of ≤0.3dB and an isolation of ≥55dB. Its common port is aligned with the center of the sampling cavity through a fiber collimator. The first inlet is connected to the optical signal transmitter, and the second inlet is connected to the optical signal decoder. After the reflected light path is returned by the total reflector, it is transmitted to the optical signal decoder through the second inlet of the circulator in a direction manner to avoid crosstalk with the transmitting light path.
[0087] The optical signal decoder includes: an optical signal receiving unit, a spectrum reading unit, and a particle concentration calculation unit. The optical signal receiving unit is used to receive optical signals. The spectrum reading unit is used to read the received optical signals to obtain spectral information. The particle concentration calculation unit, based on a machine learning algorithm, converts the spectral information into particle concentration.
[0088] The optical signal transmitter sequentially emits detection lights of different wavelengths (for example, λ1 = 532nm, λ2 = 785nm, λ3 = 1064nm) according to a preset sequence. The reflected light enters the optical signal receiving unit through the circulator, generating an original spectral data matrix I(λ, E), where λ is the wavelength and E is the spectral intensity. As Figure 2 shown, Figure 2 shows the relationship between wavelength and light intensity in electrolytes with different refractive indices.
[0089] Furthermore, the particle concentration calculation unit is built-in with a convolutional network model and a support vector machine model. Among them, the convolutional network model is used to convert spectral information into refractive index. The support vector machine model is used to convert the refractive index and external information into particle concentration.
[0090] In solutions such as electrolytes, the refractive index can accurately reflect the particle concentration, and its change can be detected by monitoring the reflected optical signal. Based on this principle, this solution constructs a dual-model analysis system, and finally realizes the accurate calculation of particle concentration by correlating the change of refractive index with the optical response characteristics.
[0091] Furthermore, the wavelength in the spectral information is used as time-domain information, and the spectral information is converted into frequency-spectrum information.
[0092] In the spectral information, the change of light intensity at different wavelengths is not intuitive enough, so it is impossible to accurately capture the refractive effect of electrolytes with different concentrations on light, and thus it is impossible to increase the calculation accuracy of electrolyte concentration. For this reason, it is necessary to convert the spectral information into frequency-spectrum information. In the frequency-spectrum information, the wavelengths of different components can be presented in the form of an energy spectrum, so the information dimension is higher and the calculation accuracy of the refractive index can be increased.
[0093] Furthermore, the following steps are adopted to convert the spectral information into frequency-spectrum information.
[0094] S1: Set the initial sampling rate f s , and convert the original spectral information into a discrete signal x[n], where n is the index of the sampling point.
[0095] In this solution, the wavelength in the spectral information is used as time-domain information, and then the spectral information is sampled, and thus the spectral information is transformed into frequency-spectrum information related to the time domain {wavelength}.
[0096] S2: Divide the discrete signal x[n] into m signal frames xm [n];
[0097] x m [n]=x[n+m*M], where N is the length of the signal frame, M is the overlapping degree of the signal frame, 0 < n < N, and m represents the index of the interval.
[0098] After dividing the discrete signal into signal frames, windowing operations can be performed on the signal frames, such as adding a Hamming window.
[0099] S3: Perform discrete Fourier transform on each signal frame to obtain the spectrum X m [k].
[0100]
[0101] where k represents the frequency index and e represents the natural constant;
[0102] S4: Calculate the power spectrum P m [k];
[0103] P m [k]=|X m [k]| 2 =X m [k]×X m * [k], X m * [k] is the complex conjugate of X m [k];
[0104] S5: Calculate the power spectrum: log P m [k];
[0105] log P m [k]=log(|X m [k]| 2 +∈);
[0106] where ε represents a very small value;
[0107] S6: Perform inverse Fourier transform on the power spectrum to obtain the cepstrum coefficients c m [n], arrange the cepstrum coefficients in a matrix according to time to generate spectral information;
[0108] where j 2 =-1, j is the imaginary unit, and k is the frequency index.
[0109] In the above solution, after performing Fourier transform on the spectral information, the resulting spectral information can highlight the variation law of light intensity at different periods, and then capture the influence of electrolytes with different refractive indices on the reflected light intensity, thus facilitating the capture of the spectral characteristics of electrolytes with different refractive indices. At the same time, the finally generated spectral information is a two-dimensional matrix with a higher information density. By processing it with a convolutional network model, more correlation information can be extracted from it.
[0110] Furthermore, the convolutional network model includes:
[0111] An input layer for inputting spectral information. The input layer receives a 10×12 cepstrum feature matrix and expands it to a 12×12 matrix by zero-padding to adapt to the convolutional operation.
[0112] A convolutional layer, which includes 3 parallel convolutional blocks, and the number of filters in the convolutional blocks gradually increases.
[0113] Each convolutional block includes a convolutional kernel, a filter, and a pooling layer. The 3 parallel convolutional blocks mainly differ in the number of filters, which are 32, 64, and 128 respectively. The operation of each convolutional block includes the following 4 steps:
[0114] Convolution calculation:
[0115] Z (1) =X*W (1) +b (1) ;
[0116] X: The input spectral information, W (1) : The convolutional kernel weight matrix (shape (3, 3, 1, 32), i.e., 【height, width, number of input channels, number of output channels; * two-dimensional convolutional operation, b (1) : The bias vector (length 32, one bias for each filter); Z (1) : The output feature map after convolution.
[0117] Activation function ReLU
[0118] A (1) : The activated feature map (with the same shape as Z (1) ).
[0119] max(0, x): Set negative values to zero and keep positive values.
[0120] Batch normalization (BatchNorm)
[0121]
[0122] μ: The mean of the current batch of data, σ 2 : The variance of the current batch of data, γ, β: Learnable scaling factor and offset parameter, ∈: A very small value (such as 10-5 ), to prevent division by zero.
[0123] Pooling process:
[0124]
[0125] Symbol explanation: P (1) : The feature map after pooling, where i and j are the indices of the samples.
[0126] Fully connected layer, used to receive the outputs of each convolutional layer to obtain combined features. Essentially, the fully connected layer sets different weights for each convolutional block, and then fuses the features finally output by each convolutional layer according to the weights to obtain combined features.
[0127] Output layer, which classifies the combined features into refractive index labels. The output layer flattens the feature map into a vector through the fully connected layer, the number of output nodes = the total number of categories, and the Softmax function converts the output into a probability distribution.
[0128] In this way, after a large amount of training on the convolutional network model, the final particle concentration can be calculated using the convolutional network model.
[0129] The support vector machine model calculates the particle concentration using the following steps;
[0130] Z1: Obtain the particles whose concentrations change in the electrolyte according to external information, and generate concentration labels for these particles; construct the original problem according to the number of concentration labels;
[0131] y i (wx i +b) > 1, or y i (wx i +b) < -1; i = 1, 2...n, n is an integer, n > 1, i
[0132] represents the sample index, w is the normal vector of the hyperplane, b is the intercept of the hyperplane, x i is the feature vector of the i-th sample, y i is the set of corresponding class labels. In the sample, the feature vector is the refractive index and the label is the particle concentration.
[0133] In the electrolyte, not all particle concentrations will change significantly. Usually, only the concentrations of specific metal ions participating in the electroplating reaction will change. Taking the chromium electroplating process as an example, the electrolyte mainly consists of chromic anhydride, and also contains additive ions such as sulfuric acid and fluorosilicate as catalysts. During the electrolysis process, these catalysts and additive ions will not be consumed. Under the condition of constant electrolyte volume, the concentrations of other ions except chromium ions basically remain stable.
[0134] In actual production, since the volume of the electrolyte is constant and non-chromium ions such as catalysts and additives are not consumed, their concentrations can be regarded as steady-state parameters. The goal of concentration monitoring is to track the dynamic changes of chromium ions in real time. For example, according to a certain electroplating process standard, the chromium ion concentration needs to be strictly controlled within the range of 100 g / L to 120 g / L. To ensure product quality (such as coating thickness tolerance), the monitoring system needs to set discrete concentration tags, and its accuracy is determined according to process sensitivity: for example, when the deviation of chromium ion concentration exceeds ±1 g / L, the coating thickness will deviate significantly from the designed value. Based on this, the monitoring tags can be discretely set to integer values such as 101 g / L, 102 g / L, 103 g / L... 120 g / L, without continuously monitoring fluctuations at the decimal point level (such as 101.1 g / L). This method simplifies the monitoring complexity through discrete tags while meeting the process accuracy requirements.
[0135] The reason for choosing the support vector machine in this solution is that the support vector machine is a classification problem, and this solution can simplify the particle concentration in the electrolyte into a classification problem through external information. y i Corresponds to a tag (specific particle concentration), while x i is the input refractive index.
[0136] The temperature in the external information is used to control the training of the support vector machine model. Before using the support vector machine model, a large amount of training is required. And the temperature data is used to determine the model object. That is, a specific support vector machine model needs to be trained for each temperature. In actual industrial production, to ensure product homogenization, the electrolyte temperature will be strictly monitored. Therefore, in essence, only 1 support vector machine model needs to be trained for each electrolyte production.
[0137] Z2: Introduce the Lagrangian function:
[0138] i and j represent the indices of the samples, α, α i 、α j respectively represent the Lagrange multipliers, L represents the Lagrangian function, and K represents the kernel function;
[0139] Z3: Solve the dual problem to find the optimal Lagrange multiplier α i ;
[0140]
[0141] max a w(α) represents the value of the maximized objective function w(α) with respect to the Lagrange multiplier α, and w(α) is the objective function of the dual problem corresponding to the original optimization problem;
[0142] Subject to the following constraints:
[0143]
[0144] 0 ≤ α i ≤ C, i = 1……n,
[0145] K(x i , x j ) is a kernel function used to calculate the inner product between samples, C represents the penalty function;
[0146] Z4: According to the optimal α i construct a decision function for prediction:
[0147] where sign is the label and x represents the unknown sample to be classified.
[0148] In the technical solution provided by this application, by calculating the optimal Lagrange multipliers, an optimal decision can be made for the objective function under the limitation of satisfying the constraint conditions, making the prediction results more accurate. After the model is established, the data prediction model needs to be trained.
[0149] The present invention also provides an electrolyte concentration monitoring system, including an information processor and a plurality of electrolyte concentration monitoring devices. The plurality of electrolyte concentration monitoring devices are evenly arranged at various positions in the electrolyte, and the information processor uses the average concentration of the particle concentrations measured at each position in the electrolyte as the concentration information of the particles in the electrolyte.
[0150] The above description is only some preferred embodiments of this application and an explanation of the technical principles applied. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this application is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with (but not limited to) technical features having similar functions disclosed in the embodiments of this application.
Claims
1. An electrolyte concentration monitoring device, characterized in that, Comprising: An optical signal transmitter for transmitting optical signals of different wavelengths; A sampling tube provided with a sampling cavity for placing an electrolyte; A total reflector disposed at the bottom of the sampling tube; An optical signal decoder for receiving an optical signal and analyzing the wavelength of the optical signal; Wherein, the optical signal transmitter is arranged at the end of the sampling tube and emits a detection light beam into the sampling cavity, the detection light beam passes through the electrolyte and irradiates onto the total reflector to form a reflected signal, the optical signal decoder receives the reflected signal, and calculates the concentration of corresponding particles in the electrolyte according to the wavelength shift of the reflected signal.
2. The electrolyte concentration monitoring device according to claim 1, characterized in that: An optical circulator is provided at the end of the sampling tube, the common port of the optical circulator faces the bottom of the sampling tube, and the first inlet and the second inlet of the optical circulator face the optical signal decoder and the total reflector respectively.
3. The electrolyte concentration monitoring device according to claim 2, wherein The optical signal decoder includes: An optical signal receiving unit for receiving an optical signal; A spectrum reading unit for reading the received optical signal to obtain spectrum information; A particle concentration calculation unit, based on a machine learning algorithm, converts the spectrum information into a particle concentration.
4. The electrolyte concentration monitoring device according to claim 3, characterized in that, The optical signal decoder further includes an information acquisition unit; The information acquisition unit is used to acquire the temperature of the electrolyte and the particle components of the electrolyte, and make the temperature of the electrolyte and the particle components of the electrolyte into external information, and send it to the particle concentration calculation unit.
5. The electrolyte concentration monitoring device according to claim 4, wherein The particle concentration calculation unit is built-in with a convolutional network model and a support vector machine model; Wherein, the convolutional network model is used to convert the spectrum information into a refractive index; The support vector machine model is used to convert the refractive index and external information into a particle concentration.
6. The electrolyte concentration monitoring device according to claim 5, characterized in that Taking the wavelength in the spectrum information as time-domain information, and converting the spectrum information into frequency-spectrum information.
7. The electrolyte concentration monitoring device according to claim 6, wherein, The following steps are adopted to convert the spectrum information into frequency-spectrum information; S1: Set the initial sampling rate f s , and convert the original spectral information into a discrete signal x[n], where n is the index of the sampling point; S2: Divide the discrete signal x[n] into m signal frames; x m [n] = x[n + m*M], where N is the length of the signal frame, M is the overlapping degree of the signal frames, 0 < n < N, and m represents the index of the interval; S3: Perform discrete Fourier transform on each signal frame to obtain the spectrum X m [k] of each signal frame; Wherein, k represents a frequency index, and e represents the natural constant; S4: Calculate the power spectrum P m [k] of each spectrum; P m [k] = |X m [k]| 2 = X m [k] × X m * [k], X m * [k] is the complex conjugate of X m [k]; S5: Calculate the power spectrum: log P m [k]; log P m [k] = log(|X m [k]| 2 + ∈); Wherein, ε represents a minimum value; S6: Perform the inverse Fourier transform on the power spectrum to obtain the cepstrum coefficients c m [n], and arrange the cepstrum coefficients in a matrix according to time; where j is the imaginary unit, and j 2 = -1, and k is the frequency index.
8. The electrolyte concentration monitoring device according to claim 1, wherein The convolutional network model includes: An input layer for inputting frequency-spectrum information; A convolutional layer including 3 parallel convolutional blocks, and the number of filters in the convolutional blocks gradually increases; A fully connected layer for receiving the outputs of each convolutional layer to obtain combined features; An output layer for classifying the combined features into refractive index labels.
9. The electrolyte concentration monitoring device according to claim 8, characterized in that, The support vector machine model calculates the particle concentration by the following steps; Z1: Obtain the particles whose particle concentration will change in the electrolyte according to the external information, and generate a concentration label for the particles; construct an original problem according to the number of concentration labels; y i (wx i +b) > 1, or y i (wx i +b) < -1; i = 1, 2... n, n is an integer, n > 1, i represents the sample index, w is the normal vector of the hyperplane, b is the intercept of the hyperplane, and x i is the feature vector of the i-th sample, and y i is the set of corresponding class labels. In the sample, the feature vector is the refractive index and the label is the particle concentration; Z2: Introduce the Lagrangian function: i and j represent the indices of the samples, α, α i , α j respectively represent the Lagrange multipliers, L represents the Lagrangian function, and K represents the kernel function; Z3: Solve the dual problem to find the optimal Lagrange multiplier α i ; max a w(α) represents the value of maximizing the objective function w(α) with respect to the Lagrange multiplier α, and w(α) is the objective function of the dual problem corresponding to the original optimization problem; Satisfying the following constraint conditions: 0 ≤ α i ≤ C, i = 1……n, K(x i ,x j ) is a kernel function used to calculate the inner product between samples, and C represents the penalty function; Z4: According to the optimal α i Construct a decision function for prediction: Among them, sign is the label, and x represents the unknown sample to be classified.
10. An electrolyte concentration monitoring system, characterized in that, Including: An information processor and several electrolyte concentration monitoring devices according to any one of claims 1 to 9; Several of the electrolyte concentration monitoring devices are evenly arranged at various positions in the electrolyte, and the information processor uses the average concentration of the particle concentrations measured at each position in the electrolyte as the concentration information of the particles in the electrolyte.