A condensation and sedimentation system for zinc powder purification

Abstract

The application belongs to the technical field of zinc powder purification, and provides a condensation and sedimentation system for zinc powder purification, which comprises the following steps: collecting a time sequence of original radiation intensity of a condensation chamber, performing frequency spectrum analysis through fast Fourier transform, calculating energy values of high and low frequency band frequency domain signals, solving a signal coherence factor and calculating a transmittance attenuation coefficient, constructing a metal thin film radiation compensation model based on a long short-term memory network, inputting the attenuation coefficient and the time sequence of original radiation intensity to output a corrected real temperature of the condensation chamber, expanding an inference layer of a full connection neural network structure, outputting a calibrated attenuation coefficient through the inference layer, selecting an effective attenuation coefficient input model by comparing a deviation threshold value, dynamically responding to temperature measurement nonlinear attenuation caused by zinc film deposition, guaranteeing production continuity and product quality of zinc powder purification, adapting to industrial multi-working condition operation requirements, being convenient to operate and controllable in cost.

Description

Technical Field

[0001] This invention belongs to the field of zinc powder purification technology, specifically a condensation sedimentation system for zinc powder purification. Background Technology

[0002] In zinc powder purification condensation and settling systems, the settling state inside the condensation chamber directly determines product quality and system stability. Because the condensation chamber is high-temperature, sealed, and filled with suspended fine zinc powder, the industry commonly employs a non-contact monitoring solution using a quartz window and an infrared thermometer. Real-time temperature measurement indirectly determines the settling state, providing support for the control system to adjust the cooling intensity.

[0003] In actual operation, the non-contact monitoring solution using a quartz window and infrared thermometer has a highly concealed temperature measurement deviation: the fine zinc powder suspended in the condensation chamber slowly adheres to the inner surface of the quartz window, forming a nanoscale semi-transparent metallic film that thickens over time and dynamically changes its light transmittance. This nanoscale semi-transparent metallic film selectively absorbs and reflects infrared radiation, causing a significant attenuation of the radiation energy received by the infrared thermometer, resulting in false data indicating continuous cooling. The control system misjudges excessive cooling and reduces the cooling intensity, ultimately leading to system overheating, causing zinc powder oxidation, abnormal particle size, and wall adhesion, affecting production continuity and product qualification rate.

[0004] The key challenge lies in the fact that this deviation is dynamic and nonlinear—affected by various factors such as film thickness and indoor operating conditions, the attenuation of radiation energy cannot be eliminated by a fixed correction coefficient or conventional calibration, becoming the core bottleneck restricting the reliability of non-contact monitoring.

[0005] Therefore, the present invention provides a condensation sedimentation system for zinc powder purification. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by this invention to solve its technical problem is:

[0008] One objective of this invention is to provide a condensation and sedimentation system for zinc powder purification, comprising:

[0009] The radiation intensity time series acquisition module acquires the original radiation intensity time series of the condenser chamber output by the infrared temperature sensor;

[0010] The frequency domain signal analysis and calculation module performs spectral analysis on the time series of the original radiation intensity of the condensing chamber, and calculates the energy value of the frequency domain signal in the high-frequency band and the total energy value of the frequency domain signal in the entire frequency band.

[0011] The signal coherence factor calculation module determines the signal coherence factor of the current quartz window based on the ratio of the energy value of the high-frequency band frequency domain signal to the total energy value of the full-frequency band frequency domain signal.

[0012] The transmittance attenuation coefficient calculation module compares the signal coherence factor with the reference coherence factor of the quartz window in the initial clean state and calculates the transmittance attenuation coefficient.

[0013] The metal thin film radiation compensation module is based on the long short-term memory network model. It constructs a metal thin film radiation compensation model and outputs the corrected real temperature value of the condenser chamber based on the transmittance attenuation coefficient and the time series of the original radiation intensity of the condenser chamber.

[0014] The attenuation coefficient dynamic calibration module extends the inference layer before the output layer of the metal thin film radiation compensation model, enabling the metal thin film radiation compensation model to cope with the nonlinear attenuation of the infrared temperature sensor measurement data caused by the zinc film covering the quartz window.

[0015] As a further improvement of the present invention, the specific process of obtaining the time series of the original radiation intensity of the condensing chamber output by the infrared temperature sensor is as follows:

[0016] Based on the actual working parameters of the condenser, the preset sampling frequency is calibrated. According to the preset sampling frequency, the original radiation intensity value of the condenser output by the infrared temperature sensor is collected in real time and the corresponding collection timestamp is recorded. The collection timestamp-original radiation intensity data are sorted in chronological order to form the time series of the original radiation intensity of the condenser.

[0017] As a further improvement of the present invention, the specific process of calculating the energy value of the high-frequency band frequency domain signal and the total energy value of the full-frequency band frequency domain signal is as follows:

[0018] The original radiation intensity time series of the condenser chamber is decomposed into frequency domain signals containing different frequency components by using the Fast Fourier Transform (FFT) algorithm.

[0019] Using the full operating frequency range of the infrared temperature sensor as the dividing threshold, all frequency components within the full operating frequency range are divided into full-band frequency domain signals; the frequency range corresponding to the signal distortion caused by zinc film deposition on the surface of the quartz window is determined, and using the frequency range as the dividing threshold, all frequency components within the frequency range are divided into high-frequency band frequency domain signals.

[0020] The total energy value of the full-band frequency domain signal is obtained by integrating the squares of the amplitudes of all frequency components in the full-band frequency domain signal; the energy value of the high-frequency band frequency domain signal is obtained by integrating the squares of the amplitudes of all frequency components in the high-frequency band frequency domain signal.

[0021] As a further improvement of the present invention, the specific process of determining the signal coherence factor of the current quartz window is as follows:

[0022] The energy value of the high-frequency band frequency domain signal and the total energy value of the full-frequency band frequency domain signal are obtained. The signal coherence factor is calculated in real time according to the formula: signal coherence factor = energy value of high-frequency band frequency domain signal ÷ total energy value of full-frequency band frequency domain signal. The signal coherence factor of the current quartz window is obtained.

[0023] As a further improvement of the present invention, the specific process for calculating the transmittance attenuation coefficient is as follows:

[0024] Obtain the signal coherence factor and reference coherence factor of the current quartz window. Compare the signal coherence factor of the current quartz window with the reference coherence factor. Calculate the transmittance attenuation coefficient according to the formula: transmittance attenuation coefficient = reference coherence factor ÷ current signal coherence factor.

[0025] As a further improvement of the present invention, the specific process of constructing the metal thin film radiation compensation model is as follows:

[0026] A long short-term memory network model was selected as the basic framework to construct a radiation compensation model for metal thin films.

[0027] The input features of the metal thin film radiation compensation model are the time series of the original radiation intensity of the condenser and the transmittance attenuation coefficient. The output features are the actual measured values ​​of the condenser temperature. The metal thin film radiation compensation model is constructed after training and validation.

[0028] As a further improvement of the present invention, the specific process for outputting the corrected true temperature value of the condenser chamber is as follows:

[0029] The original radiation intensity time series and transmittance attenuation coefficient of the condenser chamber output by the infrared temperature sensor are input into the metal thin film radiation compensation model, and the metal thin film radiation compensation model outputs the corrected true temperature value of the condenser chamber.

[0030] As a further improvement of the present invention, the specific process of extending the inference layer before the output layer of the metal thin film radiation compensation model is as follows:

[0031] A fully connected neural network structure was selected to construct the inference layer. The inference layer was integrated before the output layer of the metal thin film radiation compensation model and after the last hidden layer of the metal thin film radiation compensation model. The inference layer has 1-2 hidden layers, each with 32-64 neurons. The ReLU function was used as the activation function. The output layer is a single neuron, and the output value is the calibrated transmittance attenuation coefficient. The input features of the inference layer are the energy value of the high-frequency band frequency domain signal, the total energy value of the full-band frequency domain signal, and the signal coherence factor of the current quartz window. After training and validation, the construction of the inference layer was completed and integrated with the metal thin film radiation compensation model.

[0032] As a further improvement of the present invention, the specific process by which the metal thin film radiation compensation model can cope with the nonlinear attenuation of the infrared temperature sensor measurement data caused by the zinc film covering the quartz window is as follows:

[0033] The energy value of the high-frequency band frequency domain signal, the total energy value of the full-frequency band frequency domain signal, the signal coherence factor of the current quartz window, and the initial transmittance attenuation coefficient are input into the inference layer, and the calibrated transmittance attenuation coefficient is output. The difference between the calibrated transmittance attenuation coefficient and the initial transmittance attenuation coefficient is calculated to obtain the attenuation coefficient deviation value.

[0034] As a further improvement of the present invention, the specific details of enabling the metal thin film radiation compensation model to cope with the nonlinear attenuation of infrared temperature sensor measurement data caused by the zinc film covering the quartz window further include:

[0035] If the attenuation coefficient deviation is within the preset threshold range, the initial transmittance attenuation coefficient is deemed valid, and the initial transmittance attenuation coefficient and the original radiation intensity time series of the condenser are input into the metal thin film radiation compensation model.

[0036] If the attenuation coefficient deviation exceeds the preset threshold range, it is determined that the initial transmittance attenuation coefficient can no longer match the nonlinear attenuation characteristics of zinc film deposition. The transmittance attenuation coefficient after inference layer calibration and the time series of the original radiation intensity of the condensation chamber are then input into the metal thin film radiation compensation model.

[0037] The second objective of this invention is to provide a condensation sedimentation method for zinc powder purification, comprising:

[0038] Step S10: Obtain the time series of the original radiation intensity of the condensing chamber output by the infrared temperature sensor;

[0039] Step S20: Perform spectral analysis on the time series of the original radiation intensity of the condensing chamber, and calculate the energy value of the high-frequency band frequency domain signal and the total energy value of the full-frequency band frequency domain signal respectively;

[0040] Step S30: Determine the signal coherence factor of the current quartz window based on the ratio of the energy value of the high-frequency band frequency domain signal to the total energy value of the full-frequency band frequency domain signal;

[0041] Step S40: Compare the signal coherence factor with the reference coherence factor of the quartz window in the initial clean state, and calculate the transmittance attenuation coefficient.

[0042] Step S50: Based on the long short-term memory network model, a metal thin film radiation compensation model is constructed. According to the transmittance attenuation coefficient and the time series of the original radiation intensity of the condensing chamber, the corrected real temperature value of the condensing chamber is output through the metal thin film radiation compensation model.

[0043] Step S60: Extend the inference layer before the output layer of the metal thin film radiation compensation model so that the metal thin film radiation compensation model can cope with the nonlinear attenuation of the infrared temperature sensor measurement data caused by the zinc film covering the quartz window.

[0044] The beneficial effects of this invention are as follows:

[0045] 1. This invention addresses the problem of dynamic nonlinear deviation in infrared temperature measurement caused by zinc film deposition in quartz windows by combining the time-series data fitting advantages of long short-term memory networks with the dynamic calibration function of the inference layer through multi-module collaborative data processing logic. It breaks through the industry bottleneck that fixed correction coefficients and conventional calibration cannot eliminate this deviation, achieves accurate correction of condenser temperature, and effectively avoids misjudgment of the control system caused by false infrared temperature measurement data.

[0046] 2. A radiation compensation model is constructed using a long short-term memory network, which perfectly adapts to the dynamic characteristics of radiation intensity time series and has high fitting accuracy. The inference layer adopts a fully connected neural network structure, which can continuously learn online as the system runs and iteratively update the network weights in real time. It can accurately adapt to the nonlinear decay changes caused by long-term zinc film deposition, and can also flexibly adapt to various industrial operating conditions such as different temperatures in the condenser and zinc powder suspension concentration, ensuring long-term stability of calibration accuracy.

[0047] 3. This invention achieves dynamic correction of non-contact temperature measurement deviation through pure data processing, providing a new technical approach for the zinc powder purification industry to solve the monitoring problem under high-temperature and enclosed working conditions. It is both practical and innovative, and has positive reference and promotion value for the upgrading and optimization of related technologies in the industry. Attached Figure Description

[0048] The invention will now be further described with reference to the accompanying drawings.

[0049] Figure 1 This is a system module diagram of a zinc powder purification condensation sedimentation system according to the present invention;

[0050] Figure 2 This is a flowchart of the steps of a zinc powder purification condensation sedimentation method according to the present invention. Detailed Implementation

[0051] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0052] Example 1

[0053] like Figure 1 As shown in the embodiment of the present invention, a condensation and sedimentation system for zinc powder purification includes:

[0054] The radiation intensity time series acquisition module acquires the original radiation intensity time series of the condenser chamber output by the infrared temperature sensor;

[0055] In the radiation intensity time series acquisition module, the specific process of obtaining the original radiation intensity time series of the condenser chamber output by the infrared temperature sensor is as follows:

[0056] In the zinc powder purification condensation and sedimentation system, based on the actual working parameters of the condensation chamber (including the internal temperature range of the condensation chamber, the zinc powder suspension concentration, and the frequency of cooling intensity adjustment), the preset sampling frequency is calibrated (the preset sampling frequency must be no less than twice the working frequency of the infrared temperature sensor to satisfy the Nyquist sampling theorem and avoid signal aliasing), and the calibrated preset sampling frequency parameters are preset into the signal acquisition module of the control system.

[0057] Start the infrared temperature sensor and complete the self-test to ensure that the infrared temperature sensor communicates normally with the signal acquisition module of the control system and that the output signal of the infrared temperature sensor is stable.

[0058] The signal acquisition module of the control system triggers the infrared temperature sensor to detect radiation intensity at preset sampling frequencies. Each time it is triggered, the system acquires the original radiation intensity value of the condenser chamber output by the infrared temperature sensor in real time and records the acquisition timestamp of the corresponding acquisition time.

[0059] The continuously collected timestamp-raw radiation intensity data pairs are sorted sequentially according to time to form a continuous and complete time series of raw radiation intensity in the condenser chamber.

[0060] During the acquisition of the original radiation intensity time series in the condenser chamber, a preliminary data screening was performed simultaneously to filter out invalid data generated by abnormal fluctuations in the sensor (such as signal interruption or numerical changes exceeding the reasonable range), thus ensuring the validity of the time series.

[0061] The frequency domain signal analysis and calculation module performs spectral analysis on the time series of the original radiation intensity of the condensing chamber, and calculates the energy value of the high-frequency band frequency domain signal and the total energy value of the full-frequency band frequency domain signal, respectively.

[0062] In the frequency domain signal analysis and calculation module, the specific process of performing spectral analysis on the time series of the original radiation intensity of the condensing chamber and calculating the energy value of the high-frequency band frequency domain signal and the total energy value of the full-frequency band frequency domain signal is as follows:

[0063] The original radiation intensity time series of the condenser chamber, which is obtained and filtered by the radiation intensity time series acquisition module, is input into the signal processing module of the control system. The preset fast Fourier transform (FFT) algorithm is started to perform frequency domain transformation on the original radiation intensity time series, decomposing the continuous time domain radiation intensity signal into frequency domain signals containing different frequency components.

[0064] Using the complete operating frequency range of the infrared temperature sensor as the dividing threshold, all frequency components obtained after frequency domain conversion are screened, and all frequency components within the complete operating frequency range are uniformly divided into full-band frequency domain signals.

[0065] By conducting on-site measurements and calibrations of the quartz window in advance, the frequency range corresponding to the signal distortion caused by zinc film deposition on the surface of the quartz window was determined (the frequency range focuses on the high-frequency noise of the infrared signal caused by the deposition of the film layer; the frequency range can be flexibly adjusted according to the actual working conditions on site). Using the frequency range corresponding to the signal distortion as the dividing threshold, all frequency components after frequency domain conversion were screened again, and all frequency components within the frequency range corresponding to the signal distortion were uniformly divided into high-frequency band frequency domain signals.

[0066] Based on the above frequency domain transformation results and frequency band division results, the energy values ​​of the full-band frequency domain signal and the high-frequency band frequency domain signal are calculated respectively:

[0067] The total energy value of the full-band frequency domain signal is obtained by integrating the squares of the amplitudes of all frequency components in the full-band frequency domain signal.

[0068] The energy value of a high-frequency band frequency domain signal is obtained by integrating the squares of the amplitudes of all frequency components in the high-frequency band frequency domain signal.

[0069] For example, the complete effective operating frequency range of the sensor is 0Hz-50Hz. Preliminary field calibration revealed that when the quartz window surface is clean, the radiation signal is mainly concentrated in low frequencies (such as slow temperature fluctuations of 0-10Hz); however, when zinc film is deposited on the window surface, it will cause the infrared signal to be distorted and generate obvious high-frequency noise, the characteristic frequency range of which is concentrated in 35Hz-45Hz.

[0070] A 10-second time series of the raw radiation intensity of the condenser chamber was acquired. This was a continuous curve (time-domain signal) fluctuating up and down with time. An FFT algorithm was then used to transform this time series into the frequency domain. After the transformation, the original time-radiation intensity curve became a frequency-amplitude spectrum, and the signal was decomposed into different frequency components and their corresponding amplitudes, from 1Hz, 2Hz, 3Hz... up to 50Hz. Based on the complete operating frequency range of the infrared temperature sensor, the system sets the threshold to 0Hz~50Hz; extracts all frequency components within this range and divides them into full-band frequency domain signals; based on the previous calibration results of zinc film deposition in the quartz window, the threshold for the frequency range of signal distortion is set to 35Hz~45Hz; from the full-band signal, the frequency components between 35Hz and 45Hz are extracted separately and divided into high-frequency band frequency domain signals.

[0071] Based on the energy integral formula, the system calculates the energy values ​​of both separately:

[0072] Integrate the squares of the amplitudes of all frequency components in the range from 0Hz to 50Hz: If calculated ;

[0073] Integrate the squares of the amplitudes of all frequency components in the range of 35Hz to 45Hz: If calculated ;

[0074] The proportion of high-frequency energy can be further calculated. );

[0075] The signal coherence factor calculation module determines the signal coherence factor of the current quartz window based on the ratio of the energy value of the high-frequency band frequency domain signal to the total energy value of the full-frequency band frequency domain signal.

[0076] In the signal coherence factor calculation module, the specific process of determining the signal coherence factor of the current quartz window based on the ratio of the energy value of the high-frequency band signal to the total energy value of the full-frequency band signal is as follows:

[0077] The energy values ​​of the high-frequency band frequency domain signals and the total energy values ​​of the full-frequency band frequency domain signals calculated by the frequency domain signal analysis and calculation module are obtained to ensure that the energy values ​​of the high-frequency band frequency domain signals and the total energy values ​​of the full-frequency band frequency domain signals are consistent in time and without any abnormal missing data. The acquisition timestamps corresponding to the energy values ​​of the high-frequency band frequency domain signals and the total energy values ​​of the full-frequency band frequency domain signals are acquired simultaneously to ensure the time sequence correspondence with the original radiation intensity time series of the condenser in the radiation intensity time sequence acquisition module.

[0078] The signal coherence factor of the current quartz window is obtained by calculating the real-time ratio based on the formula: signal coherence factor = energy value of high-frequency band signal ÷ total energy value of full-band signal.

[0079] Furthermore, to enable those skilled in the art to more clearly understand the calculation process and technical effects of the above-mentioned signal coherence factor, the following provides a set of specific application examples and data under actual test conditions:

[0080] Test conditions and parameter settings:

[0081] Test object: Infrared thermometric quartz window in the zinc powder purification condensation sedimentation system;

[0082] Data acquisition frequency: 100Hz;

[0083] Time series window length: 5 seconds (i.e., each calculation cycle contains 500 discrete data points);

[0084] Frequency band division standard: the full frequency band is 0-50Hz, and the high frequency band (i.e. the characteristic frequency band of distortion clutter caused by zinc film deposition) is defined as 35-45Hz;

[0085] Real-time ratio calculation and result comparison:

[0086] During continuous production, energy value data and corresponding timestamps (accurate to milliseconds) from different periods are acquired synchronously and calculated in real time.

[0087] Operating Condition 1: Initial Cleaning State (1st Hour of Operation);

[0088] The total energy value of the full-band frequency domain signal ( The integral is 2500.5.

[0089] Energy value of high frequency band frequency domain signal ( The integral is 12.5.

[0090] Calculation process: Signal coherence factor ;

[0091] When the quartz window is completely clean, there is very little high-frequency noise and the signal coherence factor is only 0.005, indicating that the infrared radiation signal is transmitted well and there is no distortion. Here, K1 (0.005) can be used as the reference coherence factor for subsequent calculation of the transmittance attenuation coefficient.

[0092] Operating Condition 2: Zinc film deposition state (after 48 hours of continuous operation);

[0093] The total energy value of the full-band frequency domain signal ( The overall radiation intensity decreased due to the zinc film shielding, and the result after integration was 2150.8.

[0094] Energy value of high frequency band frequency domain signal ( The microstructure of the zinc film surface scatters and interferes with infrared light, resulting in a surge of high-frequency clutter, which is calculated to be 387.1 by integration.

[0095] Current signal coherence factor ;

[0096] With the deposition of zinc film, the signal coherence factor jumped significantly from 0.005 to 0.180;

[0097] The specific test data above demonstrates that this invention, by extracting the ratio of high-frequency energy to total energy (i.e., the signal coherence factor), successfully transforms the previously hidden and difficult-to-measure physical phenomenon of zinc film deposition on the quartz window surface into a highly sensitive, quantifiable, and interference-resistant characteristic indicator. Compared to existing technologies that rely solely on a single radiation intensity value, the test data from this approach clearly confirms its ability to accurately capture the early characteristics of nonlinear attenuation, providing reliable data support for subsequent calculations of the transmittance attenuation coefficient and correction of the actual temperature of the condenser chamber.

[0098] The transmittance attenuation coefficient calculation module compares the signal coherence factor with the reference coherence factor of the quartz window in the initial clean state to calculate the transmittance attenuation coefficient.

[0099] In the transmittance attenuation coefficient calculation module, the specific process of calculating the transmittance attenuation coefficient by comparing the signal coherence factor with the reference coherence factor of the quartz window in the initial clean state is as follows:

[0100] The signal coherence factor of the current quartz window is obtained from the signal coherence factor solving module, and the acquisition timestamp corresponding to the signal coherence factor of the current quartz window is obtained simultaneously to avoid data misalignment.

[0101] The reference coherence factor under the initial clean state is obtained. The reference coherence factor is a fixed threshold obtained through on-site calibration when the quartz window is completely clean and free of zinc film deposition. The signal coherence factor of the current quartz window is accurately compared with the reference coherence factor of the quartz window under the initial clean state. The transmittance attenuation coefficient is calculated in real time according to the formula: transmittance attenuation coefficient = reference coherence factor ÷ current signal coherence factor.

[0102] For example, to verify the accuracy and technical effectiveness of the above-mentioned transmittance attenuation coefficient calculation module in actual working conditions, the following provides a set of specific application examples and comparative test data of the quartz window in the condenser chamber under different operating cycles:

[0103] In the condensation and settling system of a zinc powder purification production line, the standard operating temperature inside the condensation chamber is set at 500℃. When the equipment is initially started (the quartz window is brand new and completely clean after being wiped with anhydrous ethanol), the system continuously collects data for 10 minutes. After calculation by the frequency domain signal analysis and calculation module and the signal coherence factor solving module, the fixed threshold of the reference coherence factor under the initial clean state is found to be 0.015. At this time, the true physical transmittance of the quartz window is measured to be 1.00 using a high-precision offline transmittance meter.

[0104] As production continues, zinc powder vapor gradually condenses on the inside of the quartz window, forming a zinc film of uneven thickness. The system acquires the current signal coherence factor in real time and calculates it using the formula (transmittance attenuation coefficient = reference coherence factor ÷ current signal coherence factor). Simultaneously, at corresponding time points, the system stops and uses an offline transmittance meter to measure the actual physical transmittance as a verification standard.

[0105] Running for 24 hours (light deposition period):

[0106] The system calculates the current signal coherence factor to be 0.017 in real time.

[0107] The calculated transmittance attenuation coefficient is approximately 0.882, or 88.2%, calculated as 0.015 ÷ 0.017.

[0108] The actual physical transmittance measured offline was 0.875, or 87.5%. The estimation error was only +0.7%.

[0109] Running for 72 hours (moderate deposition period):

[0110] The system calculates the current signal coherence factor to be 0.031 in real time.

[0111] The calculated transmittance attenuation coefficient is approximately 0.484, or 48.4%, calculated as 0.015 ÷ 0.031.

[0112] The actual physical transmittance measured offline was 49.1%, with an estimation error of only -0.7%.

[0113] Running for 120 hours (during heavy deposition):

[0114] The zinc film is already quite thick, and high-frequency clutter has increased significantly. The current signal coherence factor calculated by the system jumps to 0.088.

[0115] The calculated transmittance attenuation coefficient is approximately 0.170 (0.015 ÷ 0.088).

[0116] The actual physical transmittance measured offline was 16.5%, with an estimation error of only +0.5.

[0117] Compared with existing conventional technologies, the traditional linear empirical attenuation model based on running time is adopted, which sets the transmittance to decrease by a fixed 0.15 per day;

[0118] After 120 hours (5 days) of operation, the transmittance calculated by the comparative model according to the empirical model was 100% - (15% × 5) = 25.0%; however, the actual physical transmittance had already dropped nonlinearly to 16.5%, with the comparative model showing an error as high as +8.5%. If this erroneous attenuation rate were substituted into the temperature measurement system, the temperature displayed in the condenser chamber would be far lower than the actual temperature, which could easily lead to misjudgment by the control system and production accidents such as overheating and boiling of the zinc liquid. In contrast, the transmittance attenuation coefficient (17.0%) obtained by the application example of this invention through real-time ratio calculation is highly consistent with the true physical transmittance (16.5%), with the error always controlled within 1%.

[0119] The above test data fully demonstrates that the transmittance attenuation coefficient calculation logic constructed using the high-low frequency energy ratio (coherence factor) of this invention can accurately capture the nonlinear optical shielding effect caused by zinc film deposition. The attenuation coefficient output by this module has extremely high reliability, providing accurate input characteristics for subsequent metal thin film radiation compensation models, and fundamentally solving the problem of highly concealed temperature measurement deviation in non-contact monitoring schemes.

[0120] The metal thin film radiation compensation module is based on the long short-term memory network model. It constructs a metal thin film radiation compensation model and outputs the corrected real temperature value of the condenser chamber based on the transmittance attenuation coefficient and the time series of the original radiation intensity of the condenser chamber.

[0121] In the metal thin film radiation compensation module, the specific process of constructing the metal thin film radiation compensation model based on the long short-term memory network model is as follows:

[0122] A long short-term memory network model was selected as the basic framework (to adapt to the dynamic correlation capture requirements of time series data and improve the model's fitting accuracy to dynamic changes in signals) to construct a metal thin film radiation compensation model.

[0123] The input features of the metal thin film radiation compensation model are selected based on the core parameters that affect the temperature measurement deviation caused by the zinc film deposition. Specifically, these parameters include: the original radiation intensity time series of the condenser obtained and screened by the radiation intensity time series acquisition module (reflecting the original state of infrared radiation), and the transmittance attenuation coefficient calculated by the transmittance attenuation coefficient estimation module.

[0124] The output characteristic of the metal thin film radiation compensation model is clearly defined as the actual temperature value of the condenser obtained from the field experiment calibration, which is recorded as the actual measured temperature value of the condenser (as label data for model training to ensure that the model output is consistent with the actual temperature).

[0125] In the process of constructing the metal thin film radiation compensation model, the above-mentioned input feature data and corresponding real temperature calibration data under different zinc film thicknesses and different condenser temperatures were first collected through field multi-condition experiments, and then the training set, validation set and test set were divided.

[0126] The training set data is then input into a temporal neural network, and the network weights, learning rate, and other parameters are iteratively adjusted through the backpropagation algorithm to minimize the error between the model's predicted temperature and the actual temperature calibrated on-site. Subsequently, the model's hyperparameters are optimized using the validation set data to avoid overfitting or underfitting.

[0127] Finally, the model accuracy was verified through test set data to ensure that the prediction error of the metal thin film radiation compensation model was controlled within the allowable range of the condenser temperature control. The metal thin film radiation compensation model was then constructed and pre-stored in the control system.

[0128] For example, to further illustrate the specific implementation of the above-mentioned metal thin film radiation compensation model based on Long Short-Term Memory (LSTM) networks and its technical effects, detailed field test examples, model parameters, and comparative test data are provided below:

[0129] Experimental data acquisition and model hyperparameter setting:

[0130] During three months of continuous operation of a zinc powder purification condensation and sedimentation system, data was collected synchronously using high-precision thermocouples (serving as calibration benchmarks for obtaining actual measured values ​​of the condensation chamber temperature) and infrared temperature sensors deployed on-site. A total of 10,000 sets of valid time-series data samples were collected, covering different zinc film thicknesses (from clean to heavily deposited) and different actual condensation chamber temperatures (450℃-550℃). These data were divided into a training set (7,000 sets), a validation set (2,000 sets), and a test set (1,000 sets) in a 7:2:1 ratio.

[0131] Model hyperparameter settings: Network structure: Input layer + 2 hidden layers (LSTM layers) + 1 fully connected layer + output layer. Number of hidden layer neurons: The first LSTM layer is set to 64 nodes, and the second LSTM layer is set to 32 nodes (to fully extract temporal features and prevent overfitting). Optimizer selection: Adam optimizer; initial learning rate is set to 0.001. Loss function: Mean squared error (MSE); number of iterations (Epochs) is set to 200.

[0132] A segment of time-series data under complex operating conditions with heavy zinc film deposition and dynamic temperature fluctuations was randomly selected from the test set for verification. The actual temperature was calibrated on-site: the actual physical temperature inside the condenser chamber stabilized at 500.0℃. Due to the severe obstruction by the zinc film, the original radiation intensity time series output by the infrared sensor was significantly attenuated, with the corresponding uncompensated apparent temperature being only 415.3℃. The transmittance attenuation coefficient calculation module calculated the input transmittance attenuation coefficient in real time to be 0.42 (i.e., the transmittance was only 42%).

[0133] The model output of this invention is as follows: Inputting the above features into the trained LSTM metal thin film radiation compensation model, the corrected true temperature of the condenser chamber output by the model is 498.6℃. The absolute error is only -1.4℃, and the relative error is 0.28%, which is completely controlled within the allowable error range of ±2.0℃ for condenser chamber temperature control.

[0134] To verify the necessity and superiority of using the LSTM model and specific input features in this invention, the following three comparative tests were conducted using the same test set data (real temperature 500.0℃):

[0135] Comparative Example 1 (Existing Conventional Non-Contact Temperature Measurement): Without any compensation, the raw data from the infrared sensor is read directly. Test Result: The temperature reading is 415.3℃. Technical Defect: The error is as high as -84.7℃, which may cause the control system to misjudge that the condenser temperature is too low and blindly heat it, causing the zinc liquid to boil or even damage the equipment.

[0136] Comparative Example 2 (Traditional Static Emissivity Compensation Method): A fixed empirical compensation coefficient (e.g., setting a fixed attenuation of transmittance to 60%) is used for linear amplification compensation. Test Results: The measured temperature is 462.5℃. Technical Limitations: The error is -37.5℃. Because zinc film deposition is nonlinear and dynamically changing, static compensation cannot adapt to heavy deposition conditions.

[0137] Comparative Example 3 (Ordinary BP Neural Network Model): Using the same input features as this invention, but the basic framework is replaced with an ordinary BP (backpropagation) fully connected neural network without temporal memory. Test Results: The temperature measurement output fluctuates drastically between 485.2℃ and 494.1℃, with an average value of 489.6℃. Technical Limitations: Error approximately -10.4℃. Because the BP network cannot capture the dynamic fluctuations in the radiation intensity time series (temperature inertia between different times), the output results have poor anti-interference ability and severe signal jumps.

[0138] Through comparison of the above application examples and test data from multiple sets of comparative examples, it is confirmed that the LSTM-based metal thin film radiation compensation model constructed in this invention not only solves the nonlinear shielding problem by introducing a transmittance attenuation coefficient, but also smooths high-frequency clutter interference through the timing capture capability of the LSTM network. Its temperature measurement accuracy (error ≤ 1.5℃) is far superior to existing technologies and conventional neural networks, effectively solving the highly concealed temperature measurement deviation problem in existing technologies, and providing a reliable guarantee for production continuity and product qualification rate.

[0139] In the metal thin film radiation compensation module, the specific process of obtaining the corrected true temperature value of the condenser chamber based on the transmittance attenuation coefficient and the time series of the original radiation intensity of the condenser chamber through the metal thin film radiation compensation model is as follows:

[0140] The original radiation intensity time series of the condenser chamber output by the infrared temperature sensor and the calculated transmittance attenuation coefficient are obtained. The original radiation intensity time series of the condenser chamber output by the infrared temperature sensor and the transmittance attenuation coefficient are input into the metal thin film radiation compensation model. The metal thin film radiation compensation model outputs the corrected true temperature value of the condenser chamber.

[0141] The attenuation coefficient dynamic calibration module extends the inference layer before the output layer of the metal thin film radiation compensation model, enabling the metal thin film radiation compensation model to cope with the nonlinear attenuation of the infrared temperature sensor measurement data caused by the zinc film covering the quartz window.

[0142] In the dynamic calibration module for the attenuation coefficient, the specific process of extending the inference layer before the output layer of the metal thin film radiation compensation model is as follows:

[0143] A fully connected neural network structure was selected to construct the inference layer (adapting to the temporal feature output characteristics of the long short-term memory network, without complex network structure, and balancing computational efficiency and feature association learning ability). This inference layer was directly integrated before the output layer of the metal thin film radiation compensation model and after the last hidden layer of the metal thin film radiation compensation model, forming a serial model structure of long short-term memory network main body + fully connected inference layer + model output layer.

[0144] The inference layer has 1-2 hidden layers, each with 32-64 neurons. The ReLU activation function is used to solve the gradient vanishing problem in feature association learning. The output layer is a single neuron, and the output value is the calibrated transmittance attenuation coefficient.

[0145] The input features of the inference layer are determined as the energy value of the high-frequency band frequency domain signal calculated by the frequency domain signal analysis and calculation module, the total energy value of the full-band frequency domain signal calculated by the frequency domain signal analysis and calculation module, and the signal coherence factor of the current quartz window determined by the signal coherence factor solving module.

[0146] Through on-site multi-condition experiments, the input feature data of the inference layer under different zinc film thicknesses, different condenser temperatures, and different zinc powder suspension concentrations were collected. At the same time, the actual transmittance attenuation coefficient under the corresponding conditions was calibrated (as label data for inference layer training).

[0147] The collected dataset was divided into training, validation and test sets in a ratio of 7:2:1. The training set data was input into the inference layer. The network weights and biases of the inference layer were iteratively adjusted through the backpropagation algorithm to minimize the error between the calibrated transmittance attenuation coefficient output by the inference layer and the actual transmittance attenuation coefficient calibrated on site.

[0148] Optimize the hyperparameters of the inference layer (number of hidden layers, number of neurons, learning rate) using validation set data to avoid overfitting or underfitting in the inference layer;

[0149] Finally, the calibration accuracy of the inference layer was verified using test set data to ensure that the relative error between the calibrated transmittance attenuation coefficient output by the inference layer and the true value was controlled within 5%, thus completing the construction of the inference layer and integrating it with the metal thin film radiation compensation model.

[0150] For example, to verify the compensation capability of the attenuation coefficient dynamic calibration module in the face of complex operating conditions (such as high suspension concentration and nonlinear rapid film formation), the following provides a set of detailed inference layer operating parameters and comparative test data:

[0151] Example of specific structural parameters for the inference layer: Structure configuration: Two fully connected hidden layers are selected. The first hidden layer has 64 neurons, and the second hidden layer has 32 neurons. Activation function: ReLU function is used in all hidden layers, and Sigmoid function is used in the output layer (to ensure that the calibrated transmittance attenuation coefficient of the output is within a reasonable physical range of 0 to 1). Training environment: The learning rate is set to 0.0005, and mean squared error (MSE) is used as the loss function. Training is performed for 500 iterations until convergence.

[0152] Dynamic calibration process test data (severe film formation condition): When the zinc powder purification system enters the high-load operation stage (zinc powder suspension concentration is 1.5 times the calibration value), due to the rapid deposition of zinc powder, the zinc film formed on the quartz window surface exhibits complex microscopic particle stacking, resulting in extremely strong nonlinear characteristics in the attenuation of infrared light: Real-time input characteristic data: Total energy value across the entire frequency band ( ): 1850.4; High-frequency band energy value ( ): 425.5; Current signal coherence factor ( ): 0.230; Reference coherence factor ( ): 0.015;

[0153] Compare the calculation results:

[0154] Calculated derived value: The initial transmittance attenuation coefficient is calculated by the transmittance attenuation coefficient calculation module using a formula. ;

[0155] Inference layer calibration value: The above , , The input is the constructed inference layer, and the output is the calibrated transmittance attenuation coefficient. On-site calibration of the true value: the true transmittance attenuation coefficient measured synchronously using offline precision optical instruments. .

[0156] Error Analysis and Decision Logic Verification: Initial Deviation: The relative error between the initial derived value and the true value is |(0.0652−0.0575)÷0.0575|≈13.4%. Post-Calibration Deviation: The relative error between the inference layer output value and the true value is |(0.0580−0.0575)÷0.0575|≈0.87%.

[0157] Threshold determination: The deviation determination threshold is set to ±0.005, because... The value has exceeded the preset threshold. The system automatically determines that the initial derivation value is invalid and switches to 0.0580, the output of the inference layer, as the final input to the metal thin film radiation compensation model.

[0158] Comparison of scale settings and technical effects: To demonstrate the contribution of dynamic calibration to the final temperature measurement accuracy, the following comparison is set:

[0159] Comparative Example (No Dynamic Calibration): Substituting the initial derived value of 0.0652 directly into the compensation model, the final corrected temperature output is 482.5℃. Application Example (Dynamic Calibration of this Invention): Substituting the calibration value of 0.0580 into the compensation model, the final corrected temperature output is 499.2℃. Real Temperature Comparison: The actual temperature measured by the thermocouple on-site is 500.0℃.

[0160] The data comparison above shows that under complex operating conditions, the simple physical formula derivation (initial derived values) will produce a large deviation (error up to 17.5℃) due to neglecting the nonlinear scattering of the zinc film's microstructure. This invention, by extending the inference layer before the LSTM model for dynamic calibration, successfully reduced the transmittance error from 13.4% to 0.87%, thereby reducing the final temperature measurement error from 17.5℃ to 0.8℃. Test data clearly confirms that this module can effectively address the spurious data problem caused by nonlinear attenuation, significantly improving the system's measurement robustness.

[0161] In the dynamic calibration module for the attenuation coefficient, the specific process by which the metal thin-film radiation compensation model can address the nonlinear attenuation of the infrared temperature sensor measurement data caused by the zinc film covering the quartz window is as follows:

[0162] The high-frequency band frequency domain signal energy value and the total energy value of the full-band frequency domain signal are obtained in real time from the frequency domain signal analysis and calculation module. The signal coherence factor of the current quartz window is determined by the signal coherence factor solving module. The transmittance attenuation coefficient is calculated and derived by the transmittance attenuation coefficient calculation module through the formula. This is recorded as the initial transmittance attenuation coefficient. This ensures that the acquisition timestamps of all data are consistent and there is no data misalignment or missing data.

[0163] The high-frequency band energy value, the total energy value of the entire band, and the signal coherence factor are input into the constructed inference layer in real time. Based on the trained feature association relationship, the inference layer performs real-time learning and calculation on the input data and outputs the calibrated transmittance attenuation coefficient, thereby realizing the dynamic update of the transmittance attenuation coefficient.

[0164] In the control system, a threshold for judging the deviation of the transmittance attenuation coefficient is preset (this threshold is calibrated through field multi-condition experiments, and the value is the mean of the deviation between the initial transmittance attenuation coefficient and the true value ± 2 times the standard deviation). The difference between the calibrated transmittance attenuation coefficient output by the inference layer and the initial transmittance attenuation coefficient calculated and derived by the transmittance attenuation coefficient estimation module is calculated to obtain the attenuation coefficient deviation value.

[0165] The attenuation coefficient deviation value is compared with the preset deviation judgment threshold: if the attenuation coefficient deviation value is within the preset threshold range, the initial transmittance attenuation coefficient calculated and derived by the transmittance attenuation coefficient estimation module is deemed valid, and the initial transmittance attenuation coefficient and the original radiation intensity time series of the condenser chamber from the radiation intensity time series acquisition module are jointly input into the metal thin film radiation compensation model; if the attenuation coefficient deviation value exceeds the preset threshold range, the initial transmittance attenuation coefficient calculated and derived by the transmittance attenuation coefficient estimation module is deemed unable to match the nonlinear attenuation characteristics of zinc film deposition, and the calibrated transmittance attenuation coefficient output by the inference layer and the original radiation intensity time series of the condenser chamber from the radiation intensity time series acquisition module are jointly input into the metal thin film radiation compensation model.

[0166] The inference layer continuously learns online as the metal thin film radiation compensation model runs. Each time a new set of input feature data and calibration true values ​​are collected, the network weights are updated in a small increment to ensure that the calibration accuracy of the inference layer can adapt to the nonlinear decay changes caused by long-term zinc film deposition, thus achieving long-term dynamic calibration of the metal thin film radiation compensation model.

[0167] Example 2

[0168] like Figure 2As shown, based on the specific implementation process of Example 1, the present invention provides a condensation sedimentation method for zinc powder purification, comprising:

[0169] Step S10: Obtain the time series of the original radiation intensity of the condensing chamber output by the infrared temperature sensor;

[0170] Step S20: Perform spectral analysis on the time series of the original radiation intensity of the condensing chamber, and calculate the energy value of the high-frequency band frequency domain signal and the total energy value of the full-frequency band frequency domain signal respectively;

[0171] Step S30: Determine the signal coherence factor of the current quartz window based on the ratio of the energy value of the high-frequency band frequency domain signal to the total energy value of the full-frequency band frequency domain signal;

[0172] Step S40: Compare the signal coherence factor with the reference coherence factor of the quartz window in the initial clean state, and calculate the transmittance attenuation coefficient.

[0173] Step S50: Based on the long short-term memory network model, a metal thin film radiation compensation model is constructed. According to the transmittance attenuation coefficient and the time series of the original radiation intensity of the condensing chamber, the corrected real temperature value of the condensing chamber is output through the metal thin film radiation compensation model.

[0174] Step S60: Extend the inference layer before the output layer of the metal thin film radiation compensation model so that the metal thin film radiation compensation model can cope with the nonlinear attenuation of the infrared temperature sensor measurement data caused by the zinc film covering the quartz window.

[0175] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A condensation sedimentation system for zinc powder purification, characterized in that: include: The radiation intensity time series acquisition module acquires the original radiation intensity time series of the condenser chamber output by the infrared temperature sensor; The frequency domain signal analysis and calculation module performs spectral analysis on the time series of the original radiation intensity of the condensing chamber, and calculates the energy value of the frequency domain signal in the high-frequency band and the total energy value of the frequency domain signal in the entire frequency band. The signal coherence factor calculation module determines the signal coherence factor of the current quartz window based on the ratio of the energy value of the high-frequency band frequency domain signal to the total energy value of the full-frequency band frequency domain signal. The transmittance attenuation coefficient calculation module compares the signal coherence factor with the reference coherence factor of the quartz window in the initial clean state and calculates the transmittance attenuation coefficient. The metal thin film radiation compensation module is based on the long short-term memory network model. It constructs a metal thin film radiation compensation model and outputs the corrected real temperature value of the condenser chamber based on the transmittance attenuation coefficient and the time series of the original radiation intensity of the condenser chamber. The attenuation coefficient dynamic calibration module extends the inference layer before the output layer of the metal thin film radiation compensation model, enabling the metal thin film radiation compensation model to cope with the nonlinear attenuation of the infrared temperature sensor measurement data caused by the zinc film covering the quartz window.

2. The condensation and sedimentation system for zinc powder purification according to claim 1, characterized in that: The specific process for obtaining the time series of the original radiation intensity of the condenser chamber output by the infrared temperature sensor is as follows: Based on the actual working parameters of the condenser, the preset sampling frequency is calibrated. According to the preset sampling frequency, the original radiation intensity value of the condenser output by the infrared temperature sensor is collected in real time and the corresponding collection timestamp is recorded. The collection timestamp-original radiation intensity data are sorted in chronological order to form the time series of the original radiation intensity of the condenser.

3. The condensation and sedimentation system for zinc powder purification according to claim 1, characterized in that: The specific process for calculating the energy value of the high-frequency band frequency domain signal and the total energy value of the full-frequency band frequency domain signal is as follows: The original radiation intensity time series of the condenser chamber is decomposed into frequency domain signals containing different frequency components by using the Fast Fourier Transform (FFT) algorithm. Using the full operating frequency range of the infrared temperature sensor as the dividing threshold, all frequency components within the full operating frequency range are divided into full-band frequency domain signals; the frequency range corresponding to the signal distortion caused by zinc film deposition on the surface of the quartz window is determined, and using the frequency range as the dividing threshold, all frequency components within the frequency range are divided into high-frequency band frequency domain signals. The total energy value of the full-band frequency domain signal is obtained by integrating the squares of the amplitudes of all frequency components in the full-band frequency domain signal; the energy value of the high-frequency band frequency domain signal is obtained by integrating the squares of the amplitudes of all frequency components in the high-frequency band frequency domain signal.

4. The condensation and sedimentation system for zinc powder purification according to claim 1, characterized in that: The specific process for determining the signal coherence factor of the current quartz window is as follows: The energy value of the high-frequency band frequency domain signal and the total energy value of the full-frequency band frequency domain signal are obtained. The signal coherence factor is calculated in real time according to the formula: signal coherence factor = energy value of high-frequency band frequency domain signal ÷ total energy value of full-frequency band frequency domain signal. The signal coherence factor of the current quartz window is obtained.

5. The condensation and sedimentation system for zinc powder purification according to claim 1, characterized in that: The specific process for calculating the transmittance attenuation coefficient is as follows: Obtain the signal coherence factor and reference coherence factor of the current quartz window. Compare the signal coherence factor of the current quartz window with the reference coherence factor. Calculate the transmittance attenuation coefficient according to the formula: transmittance attenuation coefficient = reference coherence factor ÷ current signal coherence factor.

6. The condensation and sedimentation system for zinc powder purification according to claim 1, characterized in that: The specific process for constructing the metal thin film radiation compensation model is as follows: A long short-term memory network model was selected as the basic framework to construct a radiation compensation model for metal thin films. The input features of the metal thin film radiation compensation model are the time series of the original radiation intensity of the condenser and the transmittance attenuation coefficient. The output features are the actual measured values ​​of the condenser temperature. The metal thin film radiation compensation model is constructed after training and validation.

7. The condensation and sedimentation system for zinc powder purification according to claim 1, characterized in that: The specific process for obtaining the corrected true temperature value of the condenser chamber is as follows: The original radiation intensity time series and transmittance attenuation coefficient of the condenser chamber output by the infrared temperature sensor are input into the metal thin film radiation compensation model, and the metal thin film radiation compensation model outputs the corrected true temperature value of the condenser chamber.

8. The condensation and sedimentation system for zinc powder purification according to claim 1, characterized in that: The specific process of extending the inference layer before the output layer of the metal thin film radiation compensation model is as follows: A fully connected neural network structure was selected to construct the inference layer. The inference layer was integrated before the output layer of the metal thin film radiation compensation model and after the last hidden layer of the metal thin film radiation compensation model. The inference layer has 1-2 hidden layers, each with 32-64 neurons. The ReLU function was used as the activation function. The output layer is a single neuron, and the output value is the calibrated transmittance attenuation coefficient. The input features of the inference layer are the energy value of the high-frequency band frequency domain signal, the total energy value of the full-band frequency domain signal, and the signal coherence factor of the current quartz window. After training and validation, the construction of the inference layer was completed and integrated with the metal thin film radiation compensation model.

9. A condensation and sedimentation system for zinc powder purification according to claim 1, characterized in that: The specific process by which the metal thin film radiation compensation model can cope with the nonlinear attenuation of infrared temperature sensor measurement data caused by the zinc film covering the quartz window is as follows: The energy value of the high-frequency band frequency domain signal, the total energy value of the full-frequency band frequency domain signal, the signal coherence factor of the current quartz window, and the initial transmittance attenuation coefficient are input into the inference layer, and the calibrated transmittance attenuation coefficient is output. The difference between the calibrated transmittance attenuation coefficient and the initial transmittance attenuation coefficient is calculated to obtain the attenuation coefficient deviation value.

10. A condensation and sedimentation system for zinc powder purification according to claim 9, characterized in that: The specific details of enabling the metal thin film radiation compensation model to cope with the nonlinear attenuation of infrared temperature sensor measurement data caused by the zinc film covering the quartz window also include: If the attenuation coefficient deviation is within the preset threshold range, the initial transmittance attenuation coefficient is deemed valid, and the initial transmittance attenuation coefficient and the original radiation intensity time series of the condenser are input into the metal thin film radiation compensation model. If the attenuation coefficient deviation exceeds the preset threshold range, it is determined that the initial transmittance attenuation coefficient can no longer match the nonlinear attenuation characteristics of zinc film deposition. The transmittance attenuation coefficient after inference layer calibration and the time series of the original radiation intensity of the condensation chamber are then input into the metal thin film radiation compensation model.

Images