A circuit board etching waste liquid regeneration parameter prediction method based on machine learning

Abstract

The application discloses a circuit board etching waste liquid regeneration parameter prediction method based on machine learning and concretely relates to the field of machine learning intelligent control; a sample set is constructed through data cleaning and feature extraction, and a BP neural network prediction model is trained; in actual regeneration, grid discretization is carried out according to preset parameter steps to generate candidate regeneration parameter combinations, the effects of the combinations are predicted, multi-target comprehensive evaluation is carried out, and target regeneration parameters are determined; in the execution process, real-time states and executed reagent deduction amounts are combined to carry out dynamic correction; finally, samples are updated and the model is corrected according to actual deviations; the application can overcome the hysteresis of fixed process parameters, improve parameter prediction accuracy and process stability of etching waste liquid regeneration.

Description

Technical Field

[0001] This invention relates to the field of machine learning intelligent control technology, and more specifically, to a method for predicting parameters for the regeneration of circuit board etching waste liquid based on machine learning. Background Technology

[0002] Etching in printed circuit board (PCB) manufacturing removes unnecessary copper layers from the substrate surface to form conductive lines. During etching, the concentration of copper ions, redox state, pH, and composition continuously change, leading to a decrease in the effectiveness of the etching solution. Directly discharging the waste liquid wastes copper resources and increases treatment costs and environmental pressure. Therefore, etching waste liquid is usually regenerated through oxidation, pH adjustment, component replenishment, extraction, or electrolysis to restore its performance and achieve recycling.

[0003] The regeneration effect is affected by multiple factors: for example, in actual large-scale production, the temperature difference between summer and winter in the workshop often causes the oxidation-reduction potential (ORP) under fixed dosing control to fluctuate drastically; at the same time, the copper removal amount, etching time and production load corresponding to different products are different, resulting in large differences in the composition of waste liquid, which can easily cause the etching solution to become unstable.

[0004] Current control methods largely rely on operator experience or fixed process parameters (such as oxidant, acid addition, reaction time, and temperature). However, when the composition of the waste liquid or the state of the equipment changes, fixed parameters are difficult to match actual needs, easily leading to problems such as excessive reagents, insufficient regeneration, or inappropriate reaction times. Although some automatic control systems can perform feedback adjustment based on single parameters such as copper ion concentration, pH value, or redox potential, there are nonlinear coupling relationships between the variables, and single-parameter thresholds or simple proportional control cannot accurately describe complex regeneration processes.

[0005] With the widespread adoption of online sensors, automated dosing devices, and production information systems, a wealth of historical production data has been accumulated in the etching waste liquid regeneration process. A key challenge in improving regeneration quality and resource utilization efficiency lies in how to utilize this data to establish a mapping relationship between waste liquid state and optimal regeneration parameters, and to automatically predict suitable parameters based on the actual state of the waste liquid to be treated. This requires integrating multivariate data analysis and intelligent modeling methods to achieve precise and dynamic regeneration control. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for predicting regeneration parameters of circuit board etching waste liquid based on machine learning, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] The historical regeneration data of the circuit board etching waste liquid and the current status data of the etching waste liquid to be treated are obtained. The historical regeneration data includes at least the waste liquid composition data, waste liquid physicochemical state data, regeneration equipment operation data, regeneration process parameters, and regeneration result data.

[0009] The historical regeneration data is cleaned, time-correlated, and feature-standardized, and an etching waste liquid regeneration sample set is constructed according to different initial states of waste liquid and regeneration conditions.

[0010] Based on the etching waste liquid regeneration sample set, regeneration characteristics are extracted to characterize the initial state of the waste liquid, regeneration reaction conditions, and equipment operating status, and a sample correspondence relationship is established between the regeneration characteristics, regeneration process parameters, and regeneration results.

[0011] A machine learning prediction model is trained based on the sample correspondence, enabling the machine learning prediction model to predict the regeneration effect corresponding to the combination of regeneration process parameters based on the regeneration characteristics of the etching waste liquid and the corresponding combination of regeneration process parameters.

[0012] Based on the current state data of the etching waste liquid to be treated and the preset target regeneration state, multiple candidate regeneration process parameter combinations are generated, and the regeneration effect corresponding to each candidate regeneration process parameter combination is predicted using the machine learning prediction model.

[0013] Based on the regeneration effect, a comprehensive evaluation of each candidate combination of regeneration process parameters is conducted to determine the target regeneration process parameters, and the etching waste liquid regeneration treatment is performed according to the target regeneration process parameters.

[0014] Obtain the actual regeneration results after regeneration treatment, update the etching waste liquid regeneration sample set based on the deviation between the actual regeneration results and the predicted regeneration effect, and correct the machine learning prediction model.

[0015] Preferably, the waste liquid component data includes at least one of copper ion concentration, cuprous ion concentration, chloride ion concentration, acid concentration, complexing agent concentration, and impurity ion concentration; the waste liquid physicochemical state data includes at least one of redox potential, pH value, conductivity, density, temperature, and turbidity.

[0016] Preferably, the data cleaning of the historical regeneration data includes: identifying out-of-bounds data based on a preset physical reasonable range, identifying abrupt data based on the data change rate of adjacent sampling times, compensating for missing data according to the time correlation within the same regeneration batch, and aligning the waste liquid component data, waste liquid physicochemical state data, regeneration equipment operation data, and regeneration result data according to the data collection time.

[0017] Preferably, the regeneration characteristics include waste liquid composition characteristics, waste liquid state characteristics, equipment load characteristics, environmental state characteristics, and historical change trend characteristics; wherein, the historical change trend characteristics are obtained based on the rate of change and fluctuation of at least one parameter among copper ion concentration, redox potential, pH value, and temperature within a preset time window.

[0018] Preferably, the regeneration process parameters include at least one of the following: oxidant addition amount, acid addition amount, alkali addition amount, additive addition amount, regeneration reaction temperature, regeneration reaction time, aeration intensity, stirring speed, and waste liquid circulation flow rate.

[0019] Preferably, the machine learning prediction model adopts a BP neural network; the number of input layer nodes of the BP neural network is consistent with the total dimension of the waste liquid state characteristics and candidate regeneration process parameters, and the number of output layer nodes of the BP neural network is consistent with the number of regeneration effect indicators to be predicted.

[0020] Preferably, generating multiple candidate regeneration process parameter combinations includes: determining the initial search interval for each regeneration process parameter based on the difference between the current state of the etching waste liquid to be treated and the target regeneration state; constraining the initial search interval based on the allowable operating range of the regeneration equipment and the allowable dosing range of the reagents; and discretizing the grid within the constrained search interval according to a preset parameter step size to generate the candidate regeneration process parameter combinations.

[0021] Preferably, the regeneration effect includes at least one of the following: copper ion concentration after regeneration, redox potential, pH value, etching rate, etching uniformity, reagent consumption, regeneration time, and energy consumption; a comprehensive evaluation result of the candidate regeneration process parameter combination is obtained based on the deviation between each regeneration effect index and the corresponding target value and the evaluation weight of each regeneration effect index.

[0022] Preferably, during the execution of the target regeneration process parameters, real-time status data of the etching waste liquid is acquired according to a preset sampling period, and the real-time status data is input into the machine learning prediction model to re-predict subsequent regeneration process parameters; when the deviation between the real-time status data and the predicted status exceeds a preset deviation threshold, the regeneration process parameters that have not yet been executed are dynamically corrected. The dynamic correction includes: using the real-time status data as a new input feature, while deducting the amount of reagent added and the reaction time already executed, updating the search space of candidate regeneration process parameters to the remaining adjustable equipment allowable range, and regenerating candidate regeneration process parameter combinations in the new search space and inputting them into the machine learning prediction model for re-prediction.

[0023] Preferably, updating the etching waste liquid regeneration sample set based on the deviation between the actual regeneration result and the predicted regeneration effect includes: forming new training samples from the initial state of the waste liquid after regeneration treatment, the actual regeneration process parameters, the regeneration process state, and the actual regeneration result; and determining the training weight of the new training samples based on the prediction error, so that new training samples with prediction errors greater than a preset error threshold have higher correction weights in subsequent model training.

[0024] The technical effects and advantages of this invention are as follows:

[0025] This invention utilizes etching waste liquid regeneration data accumulated during actual circuit board production to establish a machine learning model. This model learns the nonlinear relationship between copper ion concentration, pH, redox state, temperature, equipment load, and regeneration parameters, thereby improving the matching degree between regeneration parameters and the actual waste liquid state. Before actual regeneration operation, multiple candidate regeneration process parameter combinations are generated based on the state of the waste liquid to be treated, and the regeneration results corresponding to different parameter combinations are predicted to screen out parameter combinations that may cause insufficient regeneration, excessive dosage, or excessively long reaction time. Simultaneously, reagent consumption, energy consumption, regeneration time, and post-regeneration etching capacity are all incorporated into the process parameter evaluation process, ensuring that the determination of regeneration parameters balances waste liquid performance recovery with production costs and treatment efficiency. Efficiency; During the regeneration process, the remaining regeneration process parameters are re-predicted based on continuously collected status data. When the deviation between the actual state and the predicted state exceeds the set range, the subsequent process parameters are adjusted in a timely manner to reduce the control error caused by the one-time fixed dosing method. After the regeneration process is completed, the machine learning model is continuously and incrementally updated using the actual regeneration results, so that the model can gradually adapt to the differences in operating conditions caused by different circuit board products, different production batches, different raw materials, and changes in equipment performance. This reduces the dependence of the etching waste liquid regeneration process on the operator's experience, improves the consistency of the regeneration process between different production shifts and different production lines, and provides reliable data and technical support for the stable regeneration and closed-loop recycling of etching waste liquid. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0027] Figure 2 This is a schematic diagram of the process for predicting candidate regeneration parameters, optimizing parameters, and updating feedback based on a machine learning model. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Example 1

[0030] This embodiment uses acidic copper chloride etching waste liquid from circuit boards as an example to illustrate the complete implementation process of the machine learning-based circuit board etching waste liquid regeneration parameter prediction method described in this invention.

[0031] During normal circuit board etching production, data on historical reprocessing batches and current etching waste liquid to be treated are collected. Historical reprocessing data includes waste liquid composition data, waste liquid physicochemical state data, reprocessing equipment operation data, actual reprocessing process parameters, and reprocessing result data. Specifically, waste liquid composition data includes copper ion concentration, cuprous ion concentration, chloride ion concentration, and acid concentration; waste liquid physicochemical state data includes redox potential, pH value, liquid temperature, conductivity, and density; reprocessing equipment operation data includes circulating pump flow rate, aeration flow rate, stirring status, and equipment operating time; and reprocessing process parameters include oxidant dosage, acid replenishment amount, makeup liquid dosage, reaction temperature, reaction time, aeration intensity, and circulating flow rate.

[0032] A complete etching waste liquid regeneration process is defined as starting from the entry of the waste liquid into the regeneration system until the waste liquid reaches the preset etching solution recovery conditions. For each historical regeneration batch, the waste liquid state before regeneration, the operating status of the regeneration equipment, and the actual regeneration process parameters are correlated with the corresponding regeneration results to form a historical regeneration sample.

[0033] The acquired historical regeneration data underwent data cleaning. First, out-of-bounds data was identified based on the physical reasonable range of each detection parameter and equipment operating parameter. Second, abrupt data was identified based on the rate of change of data between adjacent sampling times. For missing data with a short duration, compensation was performed using adjacent valid sample values ​​within the same regeneration batch. For data missing for more than a preset time, its sample quality level was reduced or the corresponding regeneration batch was removed from the training data. Subsequently, time correlation was performed on waste liquid component data, waste liquid physicochemical state data, equipment operating data, regeneration process parameters, and regeneration results according to the data acquisition time. Data of different dimensions were standardized to construct an etching waste liquid regeneration sample set.

[0034] Regeneration characteristics were extracted from the etching waste liquid regeneration sample set. The regeneration characteristics include waste liquid composition characteristics, waste liquid state characteristics, equipment load characteristics, and historical change trend characteristics, wherein the historical change trend characteristics are obtained based on the rate of change and fluctuation of copper ion concentration, redox potential, pH value, and temperature within a preset time window.

[0035] The waste liquid regeneration characteristics and corresponding regeneration process parameters are used as input to the machine learning prediction model. At least one of the following parameters after the completion of the corresponding regeneration batch, namely copper ion concentration, redox potential, pH value, etching rate, reagent consumption, regeneration time and energy consumption, is used as the model output to train the machine learning prediction model. The trained machine learning prediction model can predict the regeneration effect corresponding to different waste liquid states and different combinations of regeneration process parameters.

[0036] When new etching waste liquid enters the regeneration system, the copper ion concentration, redox potential, pH value, temperature, and current operating status of the batch of waste liquid are obtained, and the target regeneration state is determined according to the copper ion concentration range, redox potential range, and etching performance required for production.

[0037] The initial search range for oxidant dosage, acid dosage, reaction temperature, reaction time, aeration intensity, and circulation flow rate is determined based on the difference between the current state and the target regeneration state. This search range is then constrained by the maximum dosing capacity of the dosing device, the allowable temperature range of the reaction equipment, the allowable flow rate range of the aeration equipment, and the allowable operating range of the circulation pump. Within the constrained parameter range, the parameters are discretized according to a preset parameter step size to generate multiple candidate regeneration process parameter combinations.

[0038] The current regeneration characteristics of the etching waste liquid to be treated are input together with each candidate regeneration process parameter combination into a trained machine learning prediction model to obtain the predicted regeneration effect corresponding to each candidate regeneration process parameter combination. Based on the deviation between the predicted regeneration state and the target regeneration state, and combined with reagent consumption, regeneration time, energy consumption, and regeneration stability, each candidate regeneration process parameter combination is comprehensively evaluated. From the candidate combinations that meet the equipment operation constraints and process safety constraints, the candidate regeneration process parameter combination with the best comprehensive evaluation result is determined as the target regeneration process parameter.

[0039] The regeneration system performs etching waste liquid regeneration treatment according to the target regeneration process parameters and collects real-time status data during the regeneration process according to a preset sampling period. When the deviation between the real-time status and the predicted status is within a preset range, the system continues to execute according to the current target regeneration process parameters; when the deviation exceeds a preset deviation threshold, the current real-time status is used as the new model input, while the amount of reagent added and the reaction time already executed are deducted, and candidate regeneration process parameter combinations are regenerated based on the remaining adjustable range, and the remaining regeneration process parameters are re-predicted and determined.

[0040] After completing the regeneration of this batch of etching waste liquid, the actual regeneration results are obtained and compared with the predicted regeneration effect corresponding to the machine learning prediction model. The initial state of the waste liquid, the actual regeneration process parameters, the regeneration process status, and the actual regeneration results of this batch are used to form new training samples and added to the etching waste liquid regeneration sample set. For samples whose prediction deviation exceeds the preset error threshold and which exclude sensor malfunctions and operational malfunctions, their subsequent training weights are increased, and the updated sample set is used to incrementally correct the machine learning prediction model.

[0041] Example 2

[0042] Based on the complete implementation described in Example 1, this example further explains the construction method of the historical change trend feature in the regeneration feature. Except for the following special description, the other data acquisition, sample construction, machine learning model training, candidate regeneration parameter generation, regeneration effect prediction, target parameter determination, regeneration execution and model update process are the same as in Example 1.

[0043] Using only the state of the waste liquid at a single sampling moment may not fully reflect the trend of waste liquid change. Therefore, this embodiment further constructs trend features based on data from multiple recent sampling periods.

[0044] For example, within a time window, the rate of change of copper ion concentration is calculated:

[0045]

[0046] in, This represents the rate of change in copper ion concentration. This indicates the preset time interval.

[0047] Similarly, the rate of change of redox potential, temperature, and pH value is calculated.

[0048] Further calculations were performed on the degree of parameter fluctuation within the same time window to reflect the stability of the waste liquid state.

[0049] The features input into the machine learning model include not only the current copper ion concentration, acid concentration, and redox potential, but also the changing trends of the above parameters, the current etching load of the production line, the running time of the regeneration equipment, and the regeneration results of the previous batch.

[0050] This enables the model to distinguish between two waste liquid states with similar current detection parameters but different causes and directions of change.

[0051] Example 3

[0052] Based on the complete implementation described in Example 1, this example further illustrates the structure of the machine learning prediction model, the training process, and the generation and optimization process of candidate regeneration process parameters. Except for the technical contents specifically described below, all other steps are performed in accordance with Example 1.

[0053] Regenerated samples that have passed quality screening are selected from historical databases and divided into training datasets, validation datasets, and test datasets according to time sequence.

[0054] In one implementation, a backpropagation (BP) neural network is used to establish a regeneration effect prediction model. Specifically, the BP neural network includes an input layer, hidden layers, and an output layer; the number of nodes in the input layer is consistent with the total dimension of the waste liquid state characteristics and candidate regeneration process parameters; the hidden layer contains two layers with 64 and 32 nodes respectively, and the ReLU activation function is used; the number of nodes in the output layer is consistent with the number of regeneration result indicators to be predicted. The Adam optimizer is used during model training, and the initial learning rate is set to 0.001.

[0055] The waste liquid state characteristics and candidate regeneration process parameters are used as model inputs simultaneously:

[0056]

[0057] Where X represents the state of the waste liquid to be regenerated, P represents the candidate regeneration process parameters, and Y represents the predicted regeneration result.

[0058] The advantage of this structure is that the model does not directly provide a single regeneration parameter, but can predict the regeneration effect under different candidate process parameter conditions, thus facilitating further optimization and selection among multiple feasible process schemes.

[0059] The predicted output can include the predicted copper ion concentration after regeneration, the predicted redox potential, the predicted etching rate, the predicted reagent consumption, and the predicted processing time.

[0060] For continuous output parameters, mean squared error can be used as the model training loss function:

[0061]

[0062] Where N represents the number of training samples. This represents the model's predicted value. This represents the corresponding actual value.

[0063] If multiple outcome indicators need to be predicted simultaneously, the prediction errors corresponding to different outcome indicators can be calculated separately, and a joint loss function can be formed according to preset weights.

[0064] When new waste liquid to be recycled enters the system, the current status of the waste liquid is detected online.

[0065] For example, when the current copper ion concentration is detected to be as high as 145 g / L (higher than the target range of 120 g / L to 130 g / L) and the oxidation-reduction potential drops to 450 mV (lower than the normal production requirement of 520 mV to 580 mV), the system determines that the current main regeneration needs are to restore the copper ion oxidation state and adjust the effective components of the etching solution.

[0066] The system first obtains the initial reference range of regeneration process parameters based on historical similar samples.

[0067] Subsequently, the range of candidate parameters was further constrained based on the maximum allowable dosing rate, maximum aeration rate, allowable temperature range, and circulating pump operating range of the current regeneration equipment.

[0068] For example, multiple discrete candidate values ​​can be set around the reference value given by the machine learning model for the amount of oxidant added; similarly, multiple candidate values ​​can be set within the allowable range of the equipment for the reaction temperature, aeration flow rate, and circulation flow rate.

[0069] Several candidate regeneration process schemes are formed by combining parameters.

[0070] It should be noted that the candidate parameters do not need to be traversed infinitely across all parameter ranges. Instead, based on the control precision of the regeneration equipment, the parameters are discretized within the allowable space according to a preset parameter step size (e.g., the oxidant addition step size is set to 1L, and the reaction temperature step size is set to 2℃), thereby generating a limited number of candidate parameter combinations for traversal and prediction.

[0071] For each candidate combination of process parameters, the current waste liquid characteristics and the candidate parameter combination are input into the already trained machine learning model to predict the corresponding regeneration result.

[0072] To comprehensively evaluate different schemes, this embodiment constructs a regeneration process evaluation function:

[0073]

[0074] in, This indicates a deviation in regeneration quality. This indicates the evaluation value of drug consumption. This represents the regeneration time evaluation value. Indicates the energy consumption evaluation value. This indicates the regeneration stability evaluation value. to These represent the corresponding evaluation weights.

[0075] The above , , , , All values ​​are dimensionless values ​​after Min-Max standardization.

[0076] For production tasks with high product quality requirements, the weight corresponding to the quality deviation of regeneration can be increased; for continuous mass production scenarios, the weight corresponding to the regeneration time can be appropriately increased.

[0077] The candidate scheme with the optimal comprehensive evaluation function value is selected from all candidate parameter combinations that satisfy both equipment constraints and process safety constraints, and is used as the target regeneration process parameter.

[0078] Once the target regeneration process parameters are determined, it is not required to administer the full amount of reagents at once.

[0079] Example 4

[0080] Based on the implementation methods described in Examples 1 and 3, this example further explains the phased execution of the target regeneration process parameters, real-time status feedback, dynamic correction of remaining parameters, and incremental update of the model. Except for the following special descriptions, the remaining steps are the same as in Examples 1 and 3.

[0081] Once the target regeneration process parameters are determined, it is not necessary to administer the entire reagent dosage at once. In a preferred embodiment, the planned reagent dosage is divided into multiple execution stages.

[0082] After completing the first stage of dosing and reaction, the copper ion concentration, redox potential, pH value, and temperature of the waste liquid were retested.

[0083] The real-time state is compared with the stage state predicted by the machine learning model.

[0084] Define the state deviation index:

[0085]

[0086] in, This represents the j-th real-time detection parameter. Indicates the corresponding prediction parameters. This indicates the allowable range of variation for the corresponding parameter. This represents the weight of the state parameters.

[0087] When the state deviation index is less than the set threshold, continue to execute according to the original target regeneration parameters.

[0088] When the state deviation index exceeds the set threshold, the current real-time state replaces the original initial state of the waste liquid as the new input feature. At the same time, the amount of reagent added and the reaction time already executed are deducted. The candidate parameter search space is updated to the remaining adjustable equipment allowable range. Within this new search space, the candidate parameter combination is regenerated and re-input into the machine learning model to re-predict the remaining oxidant dosage, reaction time, aeration rate, and circulation flow rate.

[0089] This avoids mechanically executing the original parameters even after the waste liquid's state has undergone abnormal changes.

[0090] After the entire batch of etching waste liquid is regenerated, the regenerated etching liquid is tested.

[0091] The actual copper ion concentration, redox potential, pH value, etching rate, and actual reagent consumption after regeneration were compared with the model prediction results.

[0092] For ordinary samples that are accurately predicted, they are added to the historical regeneration database according to the normal sample weights.

[0093] For samples with large model prediction deviations, it is determined whether they belong to new operating conditions with limited prior training data.

[0094] When it is confirmed that there are no sensor or operational anomalies in the data, the batch of regenerated data is marked as a high-value correction sample and given higher training weight in the next round of model incremental training.

[0095] By continuously accumulating actual regeneration samples in the circuit board production process, the model can gradually learn the optimal regeneration rules for different circuit board products, different degrees of waste liquid aging, different equipment operating states, and different production load conditions.

[0096] In actual production, when the circuit board etching line generates waste liquid to be regenerated, the regeneration system automatically reads the test results of that batch of waste liquid.

[0097] The machine learning prediction system generates candidate regeneration parameters based on the current waste liquid composition and equipment status.

[0098] Taking a batch of waste liquid with an initial state of (copper ion concentration: 135 g / L, redox potential: 450 mV) as an example, if a traditional fixed parameter scheme is adopted, the oxidant needs to be added at a fixed rate of 20 L / h and reacted for 40 minutes, which can easily cause local over-oxidation and high reagent consumption. However, the target regeneration parameters selected by this system based on the comprehensive evaluation function are: the oxidant addition rate is dynamically adjusted to 15 L / h, and the reaction time is shortened to 25 minutes.

[0099] After implementing the target regeneration parameters, the copper ion concentration of this batch of waste liquid stabilized at 125 g / L, and the redox potential increased to 540 mV. Compared with the traditional fixed parameter scheme, the oxidant consumption per batch was reduced by about 15% while ensuring complete restoration of etching capability, and the risk of crystallization caused by excessive dosing was avoided.

[0100] After regeneration is completed, if the actual detection results are consistent with the predicted results of the target regeneration parameters, the data will be used as normal learning samples; if the actual results differ significantly, model correction samples will be generated based on the degree of deviation.

[0101] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting regeneration parameters of circuit board etching waste liquid based on machine learning, characterized in that, include: The historical regeneration data of the circuit board etching waste liquid and the current status data of the etching waste liquid to be treated are obtained. The historical regeneration data includes at least the waste liquid composition data, waste liquid physicochemical state data, regeneration equipment operation data, regeneration process parameters, and regeneration result data. The historical regeneration data is cleaned, time-correlated, and feature-standardized, and an etching waste liquid regeneration sample set is constructed according to different initial states of waste liquid and regeneration conditions. Based on the etching waste liquid regeneration sample set, regeneration characteristics are extracted to characterize the initial state of the waste liquid, regeneration reaction conditions, and equipment operating status, and a sample correspondence relationship is established between the regeneration characteristics, regeneration process parameters, and regeneration results. A machine learning prediction model is trained based on the sample correspondence, enabling the machine learning prediction model to predict the regeneration effect corresponding to the combination of regeneration process parameters based on the regeneration characteristics of the etching waste liquid and the corresponding combination of regeneration process parameters. Based on the current state data of the etching waste liquid to be treated and the preset target regeneration state, multiple candidate regeneration process parameter combinations are generated, and the regeneration effect corresponding to each candidate regeneration process parameter combination is predicted using the machine learning prediction model. Based on the regeneration effect, a comprehensive evaluation of each candidate combination of regeneration process parameters is conducted to determine the target regeneration process parameters, and the etching waste liquid regeneration treatment is performed according to the target regeneration process parameters. Obtain the actual regeneration results after regeneration treatment, update the etching waste liquid regeneration sample set based on the deviation between the actual regeneration results and the predicted regeneration effect, and correct the machine learning prediction model.

2. The method for predicting regeneration parameters of circuit board etching waste liquid based on machine learning according to claim 1, characterized in that: The waste liquid composition data includes at least one of the following: copper ion concentration, cuprous ion concentration, chloride ion concentration, acid concentration, complexing agent concentration, and impurity ion concentration; the waste liquid physicochemical state data includes at least one of the following: redox potential, pH value, conductivity, density, temperature, and turbidity.

3. The method for predicting regeneration parameters of circuit board etching waste liquid based on machine learning according to claim 1, characterized in that: Data cleaning of the historical regeneration data includes: identifying out-of-bounds data based on a preset physical reasonable range, identifying abrupt data based on the data change rate of adjacent sampling times, compensating for missing data according to the time correlation within the same regeneration batch, and aligning the waste liquid component data, waste liquid physicochemical state data, regeneration equipment operation data, and regeneration result data according to the data collection time.

4. The method for predicting regeneration parameters of circuit board etching waste liquid based on machine learning according to claim 1, characterized in that: The regeneration characteristics include waste liquid composition characteristics, waste liquid state characteristics, equipment load characteristics, environmental state characteristics, and historical change trend characteristics; wherein, the historical change trend characteristics are obtained based on the rate of change and fluctuation of at least one parameter among copper ion concentration, redox potential, pH value, and temperature within a preset time window.

5. The method for predicting regeneration parameters of circuit board etching waste liquid based on machine learning according to claim 1, characterized in that: The regeneration process parameters include at least one of the following: oxidant addition amount, acid addition amount, alkali addition amount, additive addition amount, regeneration reaction temperature, regeneration reaction time, aeration intensity, stirring speed, and waste liquid circulation flow rate.

6. The method for predicting regeneration parameters of circuit board etching waste liquid based on machine learning according to claim 1, characterized in that: The machine learning prediction model uses a BP neural network; the number of nodes in the input layer of the BP neural network is consistent with the total dimension of the waste liquid state characteristics and candidate regeneration process parameters, and the number of nodes in the output layer of the BP neural network is consistent with the number of regeneration effect indicators to be predicted.

7. The method for predicting regeneration parameters of circuit board etching waste liquid based on machine learning according to claim 1, characterized in that: Generating multiple candidate regeneration process parameter combinations includes: determining the initial search interval for each regeneration process parameter based on the difference between the current state of the etching waste liquid to be treated and the target regeneration state; constraining the initial search interval based on the allowable operating range of the regeneration equipment and the allowable dosing range of the reagents; and discretizing the grid within the constrained search interval according to a preset parameter step size to generate the candidate regeneration process parameter combinations.

8. The method for predicting regeneration parameters of circuit board etching waste liquid based on machine learning according to claim 1, characterized in that: The regeneration effect includes at least one of the following: copper ion concentration after regeneration, redox potential, pH value, etching rate, etching uniformity, reagent consumption, regeneration time, and energy consumption; a comprehensive evaluation result of the candidate regeneration process parameter combination is obtained based on the deviation between each regeneration effect index and the corresponding target value and the evaluation weight of each regeneration effect index.

9. The method for predicting regeneration parameters of circuit board etching waste liquid based on machine learning according to claim 1, characterized in that: During the execution of the target regeneration process parameters, real-time status data of the etching waste liquid is acquired according to a preset sampling period, and the real-time status data is input into the machine learning prediction model to re-predict the subsequent regeneration process parameters. When the deviation between the real-time status data and the predicted status exceeds a preset deviation threshold, the regeneration process parameters that have not yet been executed are dynamically corrected. The dynamic correction includes: using the real-time status data as a new input feature, while deducting the already executed reagent dosage and reaction time, updating the search space of candidate regeneration process parameters to the remaining adjustable equipment allowable range, and regenerating candidate regeneration process parameter combinations in the new search space and inputting them into the machine learning prediction model for re-prediction.

10. The method for predicting regeneration parameters of circuit board etching waste liquid based on machine learning according to claim 1, characterized in that: Updating the etching waste liquid regeneration sample set based on the deviation between the actual regeneration result and the predicted regeneration effect includes: forming new training samples from the initial state of the waste liquid after regeneration treatment, the actual regeneration process parameters, the regeneration process state, and the actual regeneration result; and determining the training weight of the new training samples based on the prediction error, so that new training samples with prediction errors greater than a preset error threshold have higher correction weights in subsequent model training.

Images