Optimized control system for internal combustion engine crankshaft quenching finishing process

Abstract

The application discloses an internal combustion engine crankshaft quenching finishing process optimization control system, comprising a multi-source process data acquisition module for acquiring multi-source process data and generating a standardized input sample set; a Prophet model construction module for establishing an improved Prophet model and outputting a predicted value, a predicted residual and a confidence interval; a black swan optimization module for performing global optimization on the model and updating model parameters; a prediction execution module for generating dynamic change trends of quenching current, scanning speed and cooling flow; a multi-objective constraint solving module for determining an optimal control parameter combination; and an adaptive control module for applying the optimal control parameter combination to an actual process and performing adjustment in a time rolling cycle. The application realizes intelligent prediction, parameter optimization and adaptive control of the internal combustion engine crankshaft quenching finishing process by fusing the improved Prophet model and the black swan optimization algorithm.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing and industrial process optimization control, and particularly relates to an internal combustion engine crankshaft quenching and finishing process optimization control system. BACKGROUND

[0002] The internal combustion engine crankshaft is a core component of the engine, and the quenching and finishing quality of the crankshaft directly affects the power performance, running stability and service life of the engine. In the traditional quenching and finishing process of the crankshaft, the process parameters such as quenching current, scanning speed and cooling flow are usually controlled by relying on experience formula or manual setting. Although this control method can maintain basic consistency in stable production, it is difficult to respond to process changes in time in the case of complex working conditions, material batch differences or equipment performance fluctuations, resulting in uneven surface hardness distribution of the crankshaft, large deformation, and obvious surface roughness fluctuation, which ultimately affects the machining precision and fatigue performance. The traditional process optimization process is long, relies on manual test verification, and has low utilization rate of process data, and cannot form dynamic learning and trend prediction of historical processes.

[0003] With the development of intelligent manufacturing and data-driven technology, machine learning and time series prediction models are gradually introduced into industrial production to assist process control. However, the widely used time series models (such as ARIMA, LSTM, etc.) in the prior art have problems such as insufficient prediction accuracy, unstable convergence, poor adaptability to external disturbances, etc. when dealing with systems with multivariate coupling characteristics, non-stationarity and process lag. Especially in the quenching process of the internal combustion engine crankshaft, the relationship between temperature change and parameters such as energy consumption, hardness and deformation is highly nonlinear, and the traditional model cannot simultaneously consider the coupling effects of trend items, seasonal items and exogenous process parameters, making it difficult to accurately depict the complex dynamic evolution law. The existing process optimization methods generally stay at the level of static parameter optimization, lack dynamic constraint optimization and real-time feedback mechanism based on prediction models, and cause disconnection between the model and the actual control process.

[0004] Therefore, how to provide an internal combustion engine crankshaft quenching and finishing process optimization control system is a problem that those skilled in the art need to solve. SUMMARY

[0005] One purpose of the present application is to propose an internal combustion engine crankshaft quenching and finishing process optimization control system, which comprehensively utilizes time series prediction modeling, intelligent optimization algorithm and adaptive process control technology, realizes multi-step trend prediction of quenching temperature, energy consumption, deformation, surface roughness and hardness by improving the Prophet model, and uses the black swan optimization algorithm to globally optimize the key parameters of the model, builds a closed-loop system of prediction-optimization-control integration, realizes dynamic adjustment and continuous correction of the crankshaft quenching and finishing process, and has the advantages of high prediction accuracy, fast response speed, strong process stability and outstanding adaptive optimization ability.

[0006] The internal combustion engine crankshaft quenching and finishing process optimization control system according to the embodiment of the present application comprises the following modules:

[0007] A multi-source process data acquisition module is used to acquire multi-source process data, pre-process the multi-source process data, and generate a standardized input sample set;

[0008] A Prophet model construction module is used to establish an improved Prophet model, perform multi-step prediction on key process parameters, and output predicted values, prediction residuals and confidence intervals;

[0009] A black swan optimization module is used to perform global optimization on the Prophet model, minimize prediction errors and update Prophet model parameters;

[0010] A prediction execution module is used to call the optimized Prophet model for retraining and prediction, and generate dynamic change trends of quenching current, scanning speed and cooling flow;

[0011] A multi-objective constraint solving module is used to establish a multi-objective process control function according to the prediction results, solve the comprehensive cost index under the constraint condition, and determine the optimal control parameter combination;

[0012] An adaptive control module is used to apply the optimal control parameter combination to the actual process, perform adjustment in a time rolling cycle, and real-time correct the Prophet model.

[0013] Optionally, the modules are realized through the following methods:

[0014] Multi-source process data of the internal combustion engine crankshaft during quenching and finishing is acquired, pre-processed, and formed into a standardized input sample set;

[0015] An improved Prophet model is constructed based on the standardized input sample set, trend items, seasonal items and exogenous regression variables are set, multi-step prediction of quenching temperature, energy consumption, deformation, surface roughness and hardness change trends is performed, and predicted values, prediction residuals and confidence intervals of each target parameter are output;

[0016] The black swan optimization algorithm is used to optimize the time series deformation factor, period coupling tensor order and dynamic self-evolution learning rate factor of the Prophet model, and the global search is performed through the bias jump direction adaptive mechanism, abnormal disturbance amplitude dynamic adjustment mechanism and population memory mechanism to minimize the prediction error of the Prophet model as the optimization objective, and the optimized Prophet model is obtained.

[0017] The optimized Prophet model is used to retrain and predict the standardized input sample set, and accurate process parameter trend prediction results and confidence intervals are output, and the dynamic change trend of the quenching current, scanning speed and cooling flow is determined according to the prediction results;

[0018] According to the prediction results, a multi-objective process control function is established, the hardness uniformity, surface roughness, energy consumption and production rhythm are taken as optimization objectives, and the prediction results are taken as constraint conditions for multi-objective constraint solving, a comprehensive cost function is calculated and the optimal control parameter combination is determined;

[0019] The optimal control parameter combination is applied to the internal combustion engine crankshaft quenching and finishing process, parameter adjustment is performed in a time rolling cycle, and the execution results are fed back to the multi-source process data acquisition link in real time, realizing continuous online correction of the Prophet model and adaptive optimization control of the process.

[0020] Optionally, the multi-source process data includes quenching coil current, frequency, scanning speed, cooling medium temperature, cooling flow, spraying timing, power signal, energy consumption information, crankshaft geometric deformation, surface roughness and hardness detection results.

[0021] Optionally, the preprocessing of the multi-source process data includes time stamp synchronization, outlier elimination and normalization processing of the multi-source process data.

[0022] Optionally, the prediction value, prediction residual and confidence interval of each target parameter are output, including:

[0023] On the basis of the standardized input sample set, an improved Prophet model is constructed, which is composed of a time series deformation perception module, a multi-dimensional period field coupling module and a self-evolution residual correction module;

[0024] The time series deformation perception module is constructed, and the Prophet model structure parameter time series deformation factor is added. The time series deformation perception module includes a time scale transformation unit and a trend reconstruction unit. The time series deformation variable representing the non-uniform time distribution characteristics is generated through the time scale transformation unit, and then the trend change path is dynamically reconstructed according to the time series deformation variable through the trend reconstruction unit;

[0025] A multi-dimensional periodic field coupling module is constructed, and a periodic coupling order of a Prophet model super parameter is added. The multi-dimensional periodic field coupling module includes a periodic decoupling unit, a periodic correlation modeling unit, and a frequency orthogonal correction unit. The periodic decoupling unit identifies characteristic components of a current period, a rotation period, and a cooling beat period. The periodic correlation modeling unit models mutual influences between the periodic signals. The frequency correction unit corrects phase offsets caused by periodic overlaps to generate periodic response quantities.

[0026] The time sequence deformation variable and the periodic response quantity, the quenching current, the scanning speed, the cooling temperature, the cooling flow rate, the spraying time sequence, and the energy consumption data jointly constitute an extended input feature set. Through a feature fusion mechanism, dynamic weighted input of multi-source information is realized.

[0027] A self-evolution residual correction module is constructed, and a dynamic self-evolution learning rate of a Prophet model training parameter is added. The self-evolution residual correction module includes a residual monitoring unit, a learning rate adjustment unit, and a stability constraint unit. The residual monitoring unit tracks the prediction residual change trend in real time. The learning rate adjustment unit automatically increases or decreases the learning rate according to the residual change direction. The stability constraint unit suppresses the overlearning phenomenon.

[0028] The improved Prophet model is used to perform multi-step prediction on the quenching temperature, the energy consumption, the deformation, the surface roughness, and the hardness change trend. The prediction values, the prediction residuals, and the confidence intervals of the target parameters are output.

[0029] Optionally, the obtaining of the optimized Prophet model parameter includes:

[0030] A parameter set to be optimized is determined, including a time sequence deformation factor, a periodic coupling order, and a dynamic self-evolution learning rate. The value range and the integer or continuous attribute of each parameter are set. The prediction step, the verification time window, and the multi-objective error weight are determined. An fitness function is constructed with the quenching temperature error, the energy consumption error, the deformation error, the surface roughness error, and the hardness error as evaluation indexes.

[0031] In the parameter feasible region, the population individuals of the black swan optimization algorithm are initialized. Each individual is assigned a set of time sequence deformation factors, periodic coupling orders, and dynamic self-evolution learning rates. The improved Prophet model is called to output multi-step prediction results and prediction residuals in the verification time window. The fitness value of each individual is calculated according to the fitness function, and the current optimal individual is recorded.

[0032] An iterative global search process is entered, a search direction is determined according to a derivative-free difference evaluation result of the fitness function based on a bias jump direction adaptive mechanism, when a threshold condition obtained from a prediction residual size and a bias score is satisfied, a directional jump update is performed on the current individual, and a feasible region clipping and an integerization processing of the periodic coupling order parameter are performed on the updated parameters;

[0033] In each iteration, an abnormal disturbance amplitude dynamic adjustment mechanism is used, the confidence interval width of the improved Prophet model in multi-step prediction is used as an uncertainty measure, the disturbance step is increased or decreased according to the uncertainty, and an iteration decay factor is superimposed, so that the parameters fall within the preset boundary and the periodic coupling order parameter maintains a legal value;

[0034] A population memory mechanism is used to maintain an elite solution set and a diversity solution set, the memory set is updated according to the fitness and the distance between solutions, the search is terminated when the iteration upper limit or the fitness convergence condition is reached, and the optimal parameter group containing the time series deformation factor parameter, the periodic coupling order parameter and the dynamic self-evolution learning rate parameter is output.

[0035] Optionally, the training and prediction of the standardized input sample set using the optimized Prophet model include:

[0036] The improved Prophet model configured with the obtained optimized parameters is called, a sliding training window length, a prediction step and a confidence interval quantile level are set based on the standardized input sample set, and the retraining of the Prophet model is completed;

[0037] After the retraining is completed, multi-step prediction is performed on the quenching temperature, energy consumption, deformation, surface roughness and hardness, and the predicted values of each target, the corresponding prediction residual and the upper and lower limit confidence interval are output according to the prediction step;

[0038] According to the obtained confidence interval width and recent prediction residual fluctuation, a risk-adjusted prediction sequence for control decision is generated, and the future change trajectory of each target is time-series arranged;

[0039] The response sensitivity information of quenching current, scanning speed and cooling flow to each target is obtained from the retrained Prophet model, the control trend increment of quenching current, scanning speed and cooling flow in the prediction time domain is calculated by combining the target weight and the reference target value;

[0040] The obtained control trend increment and the current process setting value are synthesized to form the dynamic change trend of quenching current, scanning speed and cooling flow.

[0041] Optionally, the calculation of the comprehensive cost function and the determination of the optimal control parameter combination include:

[0042] According to the multi-step prediction results and the corresponding confidence intervals, determine the optimization target set, take the hardness dispersion, surface roughness prediction value, energy consumption prediction value and production rhythm as the optimization targets, and take the quenching current, scanning speed and cooling flow as the control parameters to be solved;

[0043] Set the allowed value range and the maximum change amplitude of the adjacent control time for the quenching current, scanning speed and cooling flow, set the length of the prediction time domain and the target weight configuration, form the decision space and evaluation criteria for solving;

[0044] In the prediction time domain, the hardness dispersion, surface roughness prediction value, energy consumption prediction value and production rhythm are weighted and summarized according to the target weight to construct a cost index for measuring the comprehensive performance, which reflects the overall balance between quality consistency, surface quality, energy efficiency and rhythm;

[0045] Combined with the upper and lower limits of the confidence interval, set the quality and geometric constraints so that the lower limit of hardness is not lower than the preset threshold, the upper limit of surface roughness is not more than the preset threshold, the upper limit of deformation is not more than the preset threshold, and the values and change amplitudes of quenching current, scanning speed and cooling flow meet the specified boundary and change constraints;

[0046] Under the premise of meeting the constraints, the cost index is solved with multi-objective constraints to obtain the optimal control parameter combination of quenching current, scanning speed and cooling flow.

[0047] Optionally, the optimal control parameter combination is applied to the quenching and finishing process of the crankshaft of the internal combustion engine, and parameter adjustment is performed in a time rolling period, including:

[0048] Read the optimal control parameter combination, generate a control execution plan containing the target set value, allowed value boundary and maximum change amplitude of the adjacent period of the quenching current, scanning speed and cooling flow, and start parameter delivery according to the preset time rolling period;

[0049] In each time rolling period, the quenching current, scanning speed and cooling flow are sequentially adjusted according to the control execution plan, and the actual set value, effective time and adjustment result are recorded;

[0050] During the parameter adjustment execution process, the measured results of quenching temperature, energy consumption, deformation, surface roughness and hardness and the actually effective quenching current, scanning speed and cooling flow are synchronously collected, time alignment and data validity verification are completed, execution result data containing information required for prediction comparison are formed, and deviation information is generated;

[0051] The execution result data is fed back to the multi-source process data acquisition link in real time, time sequence deformation variables and periodic response quantities are updated, online retraining and parameter calibration of the improved Prophet model are triggered, and updated multi-step prediction results and confidence intervals are output as parameter adjustment basis and multi-objective constraint solving input of the next time rolling period.

[0052] The beneficial effects of the present application are:

[0053] The present application establishes an intelligent optimization control system for the quenching and finishing process of the crankshaft of an internal combustion engine by deeply integrating the improved Prophet prediction model and the black swan optimization algorithm. Compared with the traditional method of relying on artificial experience to set process parameters or single time series model prediction, the present application realizes dynamic modeling and trend perception of multi-source process data. By introducing time sequence deformation perception, multi-dimensional period coupling and self-evolution residual correction mechanism into the Prophet model, the model can accurately capture the complex nonlinear relationship between quenching temperature, energy consumption, deformation, surface roughness and hardness, thereby improving prediction accuracy and stability and providing a reliable basis for subsequent control optimization.

[0054] The present application uses the black swan optimization algorithm to globally optimize the core parameters of the Prophet model, and constructs a dynamic search strategy through the deviation jump direction adaptive mechanism, abnormal disturbance amplitude dynamic adjustment mechanism and population memory mechanism, so that the model has stronger convergence ability and global search performance. The optimized Prophet model can adaptively adjust parameters according to process data under different batches and different working conditions, thereby maintaining the long-term stability and generalization ability of the prediction performance, and effectively overcoming the problems of local fast convergence, parameter solidification and model degradation of traditional optimization methods. The present application realizes the collaborative optimization of multi-dimensional indexes such as quality, energy consumption and beat through a multi-objective constraint solving algorithm, ensuring uniform quenching hardness, controllable deformation, reasonable energy consumption and balanced production efficiency.

[0055] The present application constructs a process closed-loop control system with online feedback and adaptive updating function. By adjusting the optimal control parameters in the time rolling period and feeding back the execution results to the multi-source data acquisition module, the system can continuously correct the Prophet model and realize dynamic synchronous updating of model parameters and process characteristics. The closed-loop design enables the process control to have self-learning, self-correction and adaptive ability, improves the stability and consistency of the quenching and finishing process of the crankshaft of the internal combustion engine, reduces the artificial intervention and test cost, and achieves the comprehensive technical effects of improving product quality, reducing energy consumption and realizing intelligent production control. BRIEF DESCRIPTION OF DRAWINGS

[0056] The accompanying drawings are included to provide a further understanding of the application, and are incorporated in and constitute a part of this specification, illustrate embodiments of the application, and together with the description serve to explain the principles of the application. In the drawings:

[0057] Figure 1 The structure schematic diagram of the internal combustion engine crankshaft quenching finishing process optimization control system proposed in the application is shown in the figure.

[0058] Figure 2 The flowchart of the internal combustion engine crankshaft quenching finishing process optimization control method proposed in the application is shown in the figure. DETAILED DESCRIPTION

[0059] The application will now be described in further detail with reference to the drawings. These drawings show only the essential features of the application and are therefore to be regarded only as a schematic illustration. The drawings show:

[0060] REFERENCE Figure 1 The internal combustion engine crankshaft quenching finishing process optimization control system comprises the following modules:

[0061] A multi-source process data acquisition module is configured to acquire multi-source process data, pre-process the multi-source process data, and generate a standardized input sample set.

[0062] A Prophet model construction module is configured to establish an improved Prophet model, perform multi-step prediction on process key parameters, and output predicted values, prediction residuals, and confidence intervals.

[0063] A black swan optimization module is configured to perform global optimization on the Prophet model, minimize prediction errors, and update Prophet model parameters.

[0064] A prediction execution module is configured to call the optimized Prophet model to retrain and predict, and generate dynamic change trends of quenching current, scanning speed, and cooling flow.

[0065] A multi-objective constraint solving module is configured to establish a multi-objective process control function according to the prediction results, solve a comprehensive cost index under constraint conditions, and determine an optimal control parameter combination.

[0066] An adaptive control module is configured to apply the optimal control parameter combination to an actual process, perform adjustment in a time rolling cycle, and real-time correct the Prophet model.

[0067] REFERENCE Figure 2 The internal combustion engine crankshaft quenching finishing process optimization control method comprises the following steps:

[0068] Multi-source process data of the internal combustion engine crankshaft during quenching and finishing is acquired, pre-processed, and formed into a standardized input sample set.

[0069] An improved Prophet model is constructed based on a standardized input sample set, trend items, seasonal items, and exogenous regression variables are set, and the trends of quenching temperature, energy consumption, deformation, surface roughness, and hardness change are multi-step predicted, and the predicted values, prediction residuals, and confidence intervals of each target parameter are output;

[0070] The time series deformation factor, period coupling tensor order, and dynamic self-evolution learning rate factor of the Prophet model are optimized using the black swan optimization algorithm, the global search is performed through the bias jump direction adaptive mechanism, the abnormal disturbance amplitude dynamic adjustment mechanism, and the population memory mechanism, the optimization objective is to minimize the prediction error of the Prophet model, and the optimized Prophet model is obtained;

[0071] The optimized Prophet model is used to retrain and predict the standardized input sample set, and accurate process parameter trend prediction results and confidence intervals are output, and the dynamic change trends of quenching current, scanning speed, and cooling flow are determined according to the prediction results;

[0072] According to the prediction results, a multi-objective process control function is established, the hardness uniformity, surface roughness, energy consumption, and production rhythm are taken as optimization objectives, the prediction results are taken as constraint conditions for inputting multi-objective constraint solving, a comprehensive cost function is calculated, and the optimal control parameter combination is determined;

[0073] The optimal control parameter combination is applied to the internal combustion engine crankshaft quenching and finishing process, parameter adjustment is performed in a time rolling cycle, and the execution results are fed back to the multi-source process data acquisition link in real time, realizing continuous online correction of the Prophet model and adaptive optimization control of the process.

[0074] In this embodiment, the multi-source process data includes quenching coil current, frequency, scanning speed, cooling medium temperature, cooling flow, spraying timing, power signal, energy consumption information, crankshaft geometric deformation, surface roughness, and hardness detection results.

[0075] In this embodiment, the preprocessing of the multi-source process data includes timestamp synchronization, outlier removal, and normalization of the multi-source process data.

[0076] In this embodiment, the output of the predicted values, prediction residuals, and confidence intervals of each target parameter includes:

[0077] On the basis of the standardized input sample set, an improved Prophet model is constructed, which is composed of a time series deformation perception module, a multi-dimensional period field coupling module, and a self-evolution residual correction module;

[0078] The timing deformation perception module is constructed, and a timing deformation factor of a Prophet model structure parameter is added. The timing deformation perception module includes a time scale transformation unit and a trend reconstruction unit. The time scale transformation unit generates a timing deformation variable representing a non-uniform time distribution feature. The trend reconstruction unit dynamically reconstructs a trend change path according to the timing deformation variable. In this way:

[0079] The timing deformation factor enables the Prophet model to have adaptive perception and dynamic modeling capability for non-uniform time series. By automatically adjusting the time scale according to the time rhythm change in the process, the adaptive reconstruction of the trend term is realized. When the quenching temperature, energy consumption, or hardness has a sudden change or nonlinear fluctuation, the time perception weight of the Prophet model is actively changed to improve the response sensitivity and stability of the prediction to sudden changes. The Prophet model still maintains high-precision prediction and stable convergence under complex working conditions.

[0080] The time scale transformation unit generates a timing deformation variable representing a non-uniform time distribution feature. Specifically,

[0081] The time scale transformation unit dynamically calculates the time step weight according to the time interval of adjacent sampling points. When the sampling interval is short and the process changes sharply, the time weight is automatically amplified. When the sampling interval is long and the process is stable, the time weight is reduced.

[0082] According to the change amplitude of the temperature gradient, energy consumption fluctuation rate, and deformation rate, the time step is weighted and mapped to form a timing deformation variable that can represent the strength of process changes, realizing the synchronous coupling of the time dimension and process fluctuation.

[0083] The trend reconstruction unit dynamically reconstructs a trend change path according to the timing deformation variable. Specifically,

[0084] The trend reconstruction unit dynamically adjusts the change rate and direction of the trend term according to the timing deformation variable generated by the time scale transformation unit, so that the Prophet model enhances the trend response when the process fluctuation intensifies and smooths the trend change in the stable stage.

[0085] The trend reconstruction unit maps the timing deformation variable to a trend path generation function. By dynamically updating the trend weight and smoothing coefficient, a trend path that automatically evolves with time deformation is formed, realizing adaptive fitting of complex and non-stationary process.

[0086] A multi-dimensional periodic field coupling module is constructed, and a periodic coupling order of a Prophet model super parameter is added. The multi-dimensional periodic field coupling module includes a periodic decoupling unit, a periodic correlation modeling unit, and a frequency orthogonal correction unit. The periodic decoupling unit identifies characteristic components of current periods, rotation periods, and cooling beat periods. The periodic correlation modeling unit fuses and models mutual influences between periodic signals. The frequency correction unit corrects phase shifts caused by periodic overlaps to generate periodic response quantities, wherein:

[0087] The periodic correlation modeling unit fuses and models mutual influences between periodic signals, specifically as follows:

[0088] The periodic correlation modeling unit cross-correlates quenching temperature, energy consumption, cooling flow, and scanning speed, extracts phase difference and amplitude coupling characteristics, and establishes influence mapping relationships between different periodic signals.

[0089] Through dynamic aggregation and time alignment of cross-correlation characteristics, the periodic correlation modeling unit generates multi-period coupling representations inside the Prophet model, so that the Prophet model can simultaneously capture nonlinear dependency relationships between main periods and sub-periods.

[0090] The frequency correction unit corrects phase shifts caused by periodic overlaps, specifically as follows:

[0091] The frequency correction unit compares phase differences and frequency shifts of different periodic signals, identifies phase errors caused by periodic overlaps or sampling drifts in real time, and determines correction offsets through a weighted calculation method based on phase difference integration and frequency shift ratio.

[0092] After detecting the phase shift, the frequency correction unit adjusts the periodic frequency ratio according to the real-time fluctuation amplitude, remaps and interpolates the excessively overlapped or phase-shifted signals, and maintains the complete waveform structure and accurate time rhythm of the periodic signals.

[0093] The time series deformation variables and periodic response quantities, quenching current, scanning speed, cooling temperature, cooling flow, spraying time sequence, and energy consumption data jointly constitute an extended input feature set. Through a feature fusion mechanism, a multi-source information dynamic weighted input is realized. The feature fusion mechanism is a multi-source data fusion method based on dynamic weight distribution and time-varying feature interaction modeling. Through weighted summation and normalization processing of different types of input features at time steps, adaptive fusion between feature levels is realized. The feature fusion mechanism can adjust the feature weight distribution according to the real-time process state, so that the quenching current, cooling flow, and energy consumption are weighted in the fluctuation stage, while being automatically weakened in the stable stage, forming a time-varying adaptive feature input system.

[0094] The self-evolution residual correction module is constructed, and a dynamic self-evolution learning rate of a Prophet model training parameter is newly added. The self-evolution residual correction module includes a residual monitoring unit, a learning rate adjusting unit and a stability constraint unit. The residual monitoring unit is used to track the change trend of the prediction residual in real time. The learning rate adjusting unit is used to automatically increase or decrease the learning rate according to the change direction of the residual. The stability constraint unit is used to suppress the over-learning phenomenon. Specifically,

[0095] The residual monitoring unit is used to track the change trend of the prediction residual in real time. Specifically,

[0096] The residual monitoring unit is used to track the change trend of the prediction residual in real time. Specifically,

[0097] The residual monitoring unit is used to track the change trend of the prediction residual in real time. Specifically,

[0098] The residual monitoring unit is used to track the change trend of the prediction residual in real time. Specifically,

[0099] When the residual is continuously reduced, the learning rate adjusting unit gradually reduces the learning rate to prevent overfitting. When the residual is continuously increased or the shock is enhanced, the learning rate is automatically increased to speed up the error correction speed.

[0100] The residual monitoring unit is used to track the change trend of the prediction residual in real time. Specifically,

[0101] The residual monitoring unit is used to track the change trend of the prediction residual in real time. Specifically,

[0102] The residual monitoring unit is used to track the change trend of the prediction residual in real time. Specifically,

[0103] The residual monitoring unit is used to track the change trend of the prediction residual in real time. Specifically,

[0104] The improved Prophet model is used to perform multi-step prediction on the change trends of the quenching temperature, energy consumption, deformation, surface roughness and hardness, and the prediction values, prediction residuals and confidence intervals of the target parameters are output.

[0105] In the embodiment, the optimized Prophet model parameters are obtained, including:

[0106] A parameter set to be optimized is determined, including a timing deformation factor, a period coupling order, and a dynamic self-evolution learning rate, value ranges and integer or continuous properties of each parameter are set, a prediction step, a verification time window, and multi-objective error weights are determined, and a fitness function is constructed with a weighted sum of quenching temperature error, energy consumption error, deformation error, surface roughness error, and hardness error as an evaluation index;

[0107] The population individuals of the black swan optimization algorithm are initialized in the parameter feasible region, each individual is assigned a set of timing deformation factors, period coupling orders, and dynamic self-evolution learning rates, the improved Prophet model is called to output multi-step prediction results and prediction residuals in the verification time window, the fitness value of each individual is calculated according to the fitness function, and the current optimal individual is recorded;

[0108] An iterative global search process is entered, based on a bias jump direction adaptive mechanism, the search direction is determined according to the derivative-free difference evaluation result of the fitness function, when the threshold condition obtained from the prediction residual size and the bias score is met, the directional jump update is performed on the current individual, and the feasible region clipping and integer processing of the period coupling order parameter are performed on the updated parameters, wherein the bias jump direction adaptive mechanism refers to:

[0109] The sensitivity of the current individual parameter to the error is calculated by monitoring the change trend and amplitude of the Prophet model prediction residual in real time, when the residual changes in a one-way increasing trend and reaches the bias threshold, the search direction is automatically identified as a low-optimal interval, triggering the parameter jump operation to prevent falling into a local optimum;

[0110] When the bias jump direction adaptive mechanism performs the jump, a bias weight matrix is constructed according to the historical bias distribution, the direction with the fastest bias drop is preferentially selected as the new search path, and a perturbation step is applied in the search direction, so that the individual parameters jump along the approximate direction of the error optimal gradient, realizing adaptive direction correction without derivative;

[0111] When the jumped individual exceeds the parameter feasible region or the bias does not decrease, the bias jump direction adaptive mechanism automatically performs reverse rollback and step compression, recalculates the bias gradient direction, and constrains the parameters in a reasonable interval, while the period coupling order is integer processed;

[0112] In each iteration, based on the abnormal disturbance amplitude dynamic adjustment mechanism, the confidence interval width of the improved Prophet model in multi-step prediction is used as an uncertainty measure, the perturbation step is increased or decreased according to the uncertainty size, and an iteration decay factor is added, so that the parameters fall within the preset boundary and the period coupling order parameter maintains a legal value, wherein the abnormal disturbance amplitude dynamic adjustment mechanism refers to:

[0113] The dynamic adjustment mechanism for abnormal perturbation amplitude uses the confidence interval width of the Prophet model in multi-step prediction as an uncertainty indicator. When the confidence interval expands, it is identified as a high uncertainty interval, and the perturbation amplitude is automatically increased to enhance the search exploration. When the confidence interval narrows, the perturbation amplitude is reduced to improve the local convergence accuracy.

[0114] The abnormal disturbance amplitude dynamic adjustment mechanism superimposes an iteration decay factor in each iteration, so that the disturbance amplitude gradually converges with the search process, performs boundary self-calibration on the parameter change range, and ensures that all parameters are always within the preset feasible range to avoid search divergence.

[0115] When the uncertainty of prediction continues to increase, the perturbation step size and direction are automatically adjusted by weighting to enhance the global exploration capability of abnormal regions; when the stability of prediction improves, the mechanism gradually reduces the perturbation intensity and enters the fine optimization stage, realizing a dynamic self-balancing process from exploration to convergence.

[0116] A population memory mechanism is employed to maintain both the elite solution set and the diverse solution set. The memory set is updated according to the fitness and inter-solution distance criteria. The search terminates when the iteration limit is reached or the fitness convergence condition is met. The output is an optimal parameter set containing parameters of temporal deformation factor, periodic coupling order, and dynamic self-evolutionary learning rate. The population memory mechanism refers to:

[0117] By establishing two levels of storage areas—an elite solution set and a diverse solution set—the solutions with the highest fitness and those with significant structural differences are retained respectively. This approach maintains both global search capability and local convergence stability during the optimization process, thus avoiding getting trapped in local optima.

[0118] After each iteration, the priority is updated based on the individual fitness and the Euclidean distance between solutions. Duplicate or highly similar solutions are replaced, while potentially excellent but significantly different solutions are retained. This allows the memory set to evolve dynamically during the search process, maintaining the diversity and efficient distribution of the solution space.

[0119] When the rate of change of population fitness is lower than the threshold for several consecutive generations, the population memory mechanism automatically triggers memory decay, gradually reducing the weight of historical solutions and strengthening the influence of the current best solution. When the elite set converges and the diversity set is stable, the search is terminated and the global optimal parameter set, including the temporal deformation factor, the periodic coupling order, and the dynamic self-evolutionary learning rate, is output.

[0120] In this embodiment, the step of retraining and predicting the standardized input sample set using the optimized Prophet model includes:

[0121] The improved Prophet model is retrained by calling the obtained optimized parameters and setting the sliding training window length, prediction step size and confidence interval quantile level based on the standardized input sample set.

[0122] After the retraining is completed, multi-step prediction is performed on the quenching temperature, energy consumption, deformation, surface roughness and hardness, and the predicted values, corresponding prediction residuals and upper and lower limit confidence intervals of each target are output according to the prediction step length;

[0123] According to the obtained confidence interval width and recent prediction residual fluctuation, a risk-adjusted prediction sequence for control decision is generated, and the future change trajectory of each target is sequentially arranged;

[0124] The response sensitivity information of the quenching current, scanning speed and cooling flow to each target is obtained from the retrained Prophet model, and the control trend increment of the quenching current, scanning speed and cooling flow in the prediction time domain is calculated in combination with the target weight and reference target value. The calculation of the control trend increment of the quenching current, scanning speed and cooling flow in the prediction time domain is specifically:

[0125] The response sensitivity values of the quenching current, scanning speed and cooling flow to each target are extracted from the retrained Prophet model, and the influence intensity of each parameter on the target output is standardized as a sensitivity weight, which provides a quantitative basis for trend calculation;

[0126] The sensitivity weight of each parameter is weighted and fused with the target weight and reference target value to calculate the contribution rate of each control parameter to the target deviation in the prediction time domain, and the trend adjustment direction and relative amplitude are determined;

[0127] The control trend increment of the quenching current, scanning speed and cooling flow is generated according to the weighted calculation result, and a time domain smoothing factor is introduced for continuous correction to ensure smooth transition of the trend change in the prediction period, and the control strategy output of multi-target coordinated optimization is realized;

[0128] The obtained control trend increment and the current process setting value are synthesized to form the dynamic change trend of the quenching current, scanning speed and cooling flow.

[0129] In the embodiment, the calculation of the comprehensive cost function and the determination of the optimal control parameter combination include:

[0130] According to the multi-step prediction result and the corresponding confidence interval, an optimization target set is determined, the hardness dispersion, surface roughness prediction value, energy consumption prediction value and production rhythm are taken as optimization targets, and the quenching current, scanning speed and cooling flow are taken as control parameters to be solved;

[0131] The quenching current, scanning speed and cooling flow are respectively set to have an allowed value range and a maximum change amplitude at adjacent control time, and the length of the prediction time domain and the target weight configuration are set to form a decision space and evaluation criteria for solving;

[0132] In the prediction time domain, the hardness dispersion, the surface roughness prediction value, the energy consumption prediction value and the production rhythm are weighted and aggregated according to the target weight to construct a cost index for measuring the comprehensive performance, and the cost index reflects the overall balance relationship between the quality consistency, the surface quality, the energy efficiency and the rhythm.

[0133] In combination with the upper and lower limits of the confidence interval, the quality and geometric constraints are set so that the lower limit of the hardness is not lower than the preset threshold, the upper limit of the surface roughness is not higher than the preset threshold, the upper limit of the deformation is not higher than the preset threshold, and the values and variation amplitudes of the quenching current, the scanning speed and the cooling flow meet the specified boundary and variation constraints;

[0134] Under the premise of meeting the constraint conditions, the cost index is solved by multi-objective constraint to obtain the optimal control parameter combination of the quenching current, the scanning speed and the cooling flow, wherein the cost index is solved by multi-objective constraint, specifically:

[0135] The hardness uniformity, the surface roughness, the energy consumption and the production rhythm are taken as optimization targets, a comprehensive cost function is established, and constraint conditions including device power, temperature stability and safety threshold are applied;

[0136] Under the premise of meeting the constraint conditions, the comprehensive cost function is optimized by multi-objective optimization using the NSGA-II algorithm, the diversity of solutions is maintained through non-dominated sorting and crowding distance, and the multi-objective balance of minimum energy consumption, maximum hardness uniformity and production rhythm optimization is realized;

[0137] The solutions that do not meet the constraint conditions are corrected by applying a penalty function, and the Pareto frontier solution set is selected according to the fitness ranking and constraint matching degree, and finally the solution with the optimal cost function value is selected as the optimal control parameter combination of the quenching current, the scanning speed and the cooling flow, to realize the global optimal control of process performance and energy efficiency.

[0138] In the embodiment, the optimal control parameter combination is applied to the quenching and finishing process of the crankshaft of the internal combustion engine, and parameter adjustment is performed in a time rolling period, including:

[0139] The optimal control parameter combination is read to generate a control execution plan including target set values of the quenching current, the scanning speed and the cooling flow, allowable value boundaries and maximum variation amplitudes of adjacent periods, and the parameter delivery is started according to the preset time rolling period;

[0140] In each time rolling period, the quenching current, the scanning speed and the cooling flow are sequentially adjusted according to the control execution plan, and the actual delivered set value, the effective time and the adjustment result are recorded;

[0141] During the parameter adjustment execution process, the measured results of quenching temperature, energy consumption, deformation amount, surface roughness and hardness, and the actually effective quenching current, scanning speed and cooling flow are synchronously collected, time alignment and data validity check are completed, execution result data containing information required for prediction comparison are formed, and deviation information is generated;

[0142] The execution result data and the deviation information are real-time returned to the multi-source process data collection link, the time series deformation variables and the periodic response quantities are updated, the online retraining and parameter calibration of the improved Prophet model are triggered, the updated multi-step prediction results and confidence intervals are output, which are used as the basis for parameter adjustment and input for multi-objective constraint solving in the next time rolling period.

[0143] Embodiment 1

[0144] In order to verify the feasibility of the application in implementation, the application is applied to a large automobile engine manufacturing enterprise, and the internal combustion engine crankshaft quenching finishing process optimization control method of the application is tested in practical application. The production line mainly produces 2.0L gasoline engine crankshafts, adopts medium-frequency induction heating equipment for surface quenching, and completes the final finishing through an automatic grinding system. In the past, the production line mainly relied on manual experience to adjust process parameters, resulting in large fluctuations in hardness distribution, difficulty in matching quenching current and scanning speed, lag in cooling flow control, and high overall energy consumption. During the production process, problems such as uneven surface hardness, local overburning and excessive deformation often occur, and the product scrap rate has been maintained at about 3% for a long time.

[0145] After implementing the application, the system first collects data such as temperature, energy consumption, cooling flow, deformation and surface roughness in real time through a multi-source process data collection module, with a sampling frequency of once every 2 seconds, and an average of about 2500 data records obtained for each workpiece. The collected data is standardized and input into the improved Prophet model. The model is trained based on the set trend item, seasonal item and exogenous regression variable, combined with time series deformation perception, multi-dimensional period coupling and self-evolution residual correction mechanism, and outputs multi-step prediction values and confidence intervals of each target parameter. Subsequently, the black swan optimization algorithm globally optimizes the time series deformation factor, period coupling tensor order and dynamic self-evolution learning rate factor of the Prophet model. The algorithm uses the deviation jump direction adaptive mechanism and the abnormal disturbance amplitude dynamic adjustment mechanism to improve the response capability to nonlinear process fluctuations, and maintains the stability of the parameters through the population memory mechanism, so that the model prediction error is significantly reduced.

[0146] The optimized model was continuously predicted and controlled for 7 days in the test period. When the Prophet model generates new prediction results, the system establishes a multi-objective process control function according to the predicted temperature, energy consumption, deformation, roughness and hardness trend, and takes the hardness uniformity, surface roughness, energy consumption and production rhythm as the optimization target. The system automatically solves the comprehensive cost function to generate the optimal quenching current, scanning speed and cooling flow combination, and adjusts the parameters in the time rolling period. Through the real-time feedback mechanism of the adaptive control module, the actual measurement data is continuously returned to the model end, and the Prophet model online corrects the trend item and exogenous regression weight, and gradually forms a stable prediction-optimization-control closed loop.

[0147] Table 1 Comparison of key performance of crankshaft quenching finishing process before and after optimization

[0148] Index Category Before Optimization Mean After Optimization Mean Improvement Range Data Source Period (days) Surface Hardness (HRC) 58.4 59.2 +1.4% 30 Hardness Distribution Standard Deviation (HRC) 1.1 0.4 -63.6% 30 Surface Roughness Ra (μm) 0.90 0.67 -25.6% 30 Deformation (mm) 0.038 0.026 -31.6% 30 Energy Consumption (kWh / piece) 3.42 3.03 -11.4% 30 Quenching Current Fluctuation (%) ±6.2 ±2.1 -66.1% 30 Cooling Flow Fluctuation (L / min) ±2.5 ±0.8 -68.0% 30 Production Tact Time (min / piece) 3.5 3.2 -8.6% 30 Product Pass Rate (%) 97.1 99.3 +2.2% 30

[0149] From the data analysis in Table 1, the method of the present application has achieved significant optimization effect in actual industrial application. In terms of surface hardness of crankshaft, the average value is increased from 58.4HRC to 59.2HRC, about 1.4% higher, which shows that the process control of the system after optimization is more accurate, and the hardness of the quenching layer is stable and sufficient, avoiding the overheating or overcooling phenomenon in traditional manual adjustment. The standard deviation of hardness distribution is reduced from 1.1HRC to 0.4HRC, with a reduction of 63.6%, which shows that the hardness distribution of each measuring point is more uniform, and the material performance consistency is significantly improved. The surface roughness Ra is reduced from 0.90μm to 0.67μm, about 25.6% higher, which benefits from the dynamic optimization of the prediction model for scanning speed and cooling flow, making the surface structure more fine and reducing the fine grinding allowance and subsequent finishing times.

[0150] From the energy efficiency and process stability point of view, the energy consumption is reduced from 3.42kWh / piece to 3.03kWh / piece, with an average energy saving of 11.4%. The quenching current fluctuation is reduced from ±6.2% to ±2.1%, and the cooling flow fluctuation is reduced from ±2.5L / min to ±0.8L / min, with obvious improvement in control accuracy. This shows that through the global optimization of key parameters of the Prophet model by the black swan optimization algorithm, the collaborative control of current and flow is realized, and the system can maintain the balance of energy input and cooling efficiency in the real-time fluctuation environment, thereby reducing the invalid energy consumption and improving the heat treatment consistency. The deformation is reduced from 0.038mm to 0.026mm, about 31.6% lower, which shows that this method effectively relieves the geometric deformation caused by thermal stress difference.

[0151] Finally, in terms of production efficiency and comprehensive quality, the production rhythm is shortened from 3.5 minutes to 3.2 minutes, with a rhythm optimization of 8.6%, and the whole line rhythm is more balanced. The product qualified rate is improved from 97.1% to 99.3%, and the scrap rate is significantly reduced, which reflects the industrial feasibility and stability of the prediction-optimization-control closed-loop mechanism. Through improving the trend prediction ability of the Prophet model and the global search advantage of the black swan algorithm, the invention realizes multi-objective dynamic optimization control, enabling the system to adapt to different batches of materials and equipment fluctuations, maintaining the stability and efficiency of the production process. The invention method has achieved practical and verifiable improvements in hardness uniformity, surface quality, energy efficiency control, and production rhythm, fully demonstrating its engineering application value and technical advancement in the internal combustion engine crankshaft quenching and finishing process.

[0152] The above is only the preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical scheme and inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. An optimization control system for the quenching and finishing process of an internal combustion engine crankshaft, characterized in that, Includes the following modules: The multi-source process data acquisition module is used to acquire multi-source process data, preprocess the multi-source process data, and generate a standardized input sample set; The Prophet model building module is used to build an improved Prophet model, perform multi-step predictions of key process parameters, and output predicted values, prediction residuals, and confidence intervals. The Black Swan optimization module is used to perform global optimization on the Prophet model, minimize prediction error, and update the Prophet model parameters. The prediction execution module is used to call the optimized Prophet model to retrain and predict, generating dynamic trends of quenching current, scanning speed and cooling flow rate. The multi-objective constraint solving module is used to establish multi-objective process control functions based on prediction results, solve the comprehensive cost index under constraints, and determine the optimal combination of control parameters. The adaptive control module is used to apply the optimal combination of control parameters to the actual process, perform adjustments in a rolling time cycle, and correct the Prophet model in real time.

2. A method for optimizing and controlling the quenching and finishing process of an internal combustion engine crankshaft, applied to the optimization and control system for the quenching and finishing process of an internal combustion engine crankshaft as described in claim 1, characterized in that... include: Collect multi-source process data of internal combustion engine crankshaft during quenching and finishing, preprocess the multi-source process data, and form a standardized input sample set; An improved Prophet model is constructed based on a standardized input sample set. Trend, seasonal and exogenous regression variables are set to make multi-step predictions on the trends of quenching temperature, energy consumption, deformation, surface roughness and hardness. The predicted values, prediction residuals and confidence intervals of each target parameter are output. The temporal deformation factor, periodic coupling tensor order, and dynamic self-evolutionary learning rate factor of the Prophet model are optimized using the Black Swan optimization algorithm. A global search is performed through the bias jump direction adaptive mechanism, the abnormal perturbation amplitude dynamic adjustment mechanism, and the population memory mechanism to minimize the prediction error of the Prophet model and obtain the optimized Prophet model. The optimized Prophet model is used to retrain and predict the standardized input sample set, and the accurate prediction results and confidence intervals of process parameter trends are output. Based on the prediction results, the dynamic change trends of quenching current, scanning speed and cooling flow rate are determined. Based on the prediction results, a multi-objective process control function is established, with hardness uniformity, surface roughness, energy consumption and production cycle as optimization objectives, and the prediction results as constraint inputs for multi-objective constraint solution, calculating the comprehensive cost function and determining the optimal combination of control parameters; The optimal combination of control parameters is applied to the quenching and finishing process of the internal combustion engine crankshaft. The parameters are adjusted according to the time rolling cycle, and the execution results are transmitted back to the multi-source process data acquisition link in real time, so as to realize the continuous online correction of the Prophet model and the adaptive optimization control of the process.

3. The method for optimizing and controlling the quenching and finishing process of an internal combustion engine crankshaft according to claim 2, characterized in that, The multi-source process data includes quenching coil current, frequency, scanning speed, cooling medium temperature, cooling flow rate, spraying sequence, power signal, energy consumption information, crankshaft geometric deformation, and surface roughness and hardness detection results.

4. The method for optimizing and controlling the quenching and finishing process of an internal combustion engine crankshaft according to claim 2, characterized in that, The preprocessing of multi-source process data includes timestamp synchronization, outlier removal, and normalization.

5. The method for optimizing and controlling the quenching and finishing process of an internal combustion engine crankshaft according to claim 2, characterized in that, The output includes the predicted values, prediction residuals, and confidence intervals of each target parameter, including: Based on the standardized input sample set, an improved Prophet model is constructed. The Prophet model consists of a temporal deformation sensing module, a multidimensional periodic field coupling module, and a self-evolving residual correction module. A temporal deformation perception module is constructed, and a temporal deformation factor is added as a structural parameter of the Prophet model. The temporal deformation perception module includes a time scale transformation unit and a trend reconstruction unit. The time scale transformation unit generates temporal deformation variables that characterize the non-uniform time distribution, and the trend reconstruction unit dynamically reconstructs the trend change path based on the temporal deformation variables. A multidimensional periodic field coupling module is constructed, and a new hyperparameter periodic coupling order of the Prophet model is added. The multidimensional periodic field coupling module includes a periodic decoupling unit, a periodic correlation modeling unit, and a frequency orthogonal correction unit. The periodic decoupling unit identifies the characteristic components of the current period, rotation period, and cooling beat period. The periodic correlation modeling unit performs fusion modeling on the mutual influence between each periodic signal. The frequency correction unit corrects the phase shift caused by periodic overlap and generates periodic response quantities. The time-series deformation and periodic response, together with quenching current, scanning speed, cooling temperature, cooling flow rate, spray timing and energy consumption data, constitute an extended input feature set. Through the feature fusion mechanism, dynamic weighted input of multi-source information is realized. An evolutionary residual correction module is constructed, and a dynamic evolutionary learning rate is added to the Prophet model training parameters. The evolutionary residual correction module includes a residual monitoring unit, a learning rate adjustment unit, and a stability constraint unit. The residual monitoring unit tracks and predicts the trend of residual changes in real time, and the learning rate adjustment unit automatically increases or decreases the learning rate according to the direction of residual change. At the same time, the stability constraint unit suppresses overlearning. The improved Prophet model is used to make multi-step predictions of the trends in quenching temperature, energy consumption, deformation, surface roughness and hardness, and output the predicted values, prediction residuals and confidence intervals of each target parameter.

6. The method for optimizing and controlling the quenching and finishing process of an internal combustion engine crankshaft according to claim 2, characterized in that, The process of obtaining the optimized Prophet model parameters includes: The set of parameters to be optimized is determined, including the temporal deformation factor, the periodic coupling order and the dynamic self-evolution learning rate. The value range and integer or continuous attribute of each parameter are set respectively. The prediction step size, the verification time window and the multi-objective error weight are determined. A fitness function with the weighted sum of quenching temperature error, energy consumption error, deformation error, surface roughness error and hardness error as the evaluation index is constructed. Initialize the population of Black Swan optimization algorithm individuals within the parameter feasible region, assign each individual a set of temporal deformation factors, periodic coupling order and dynamic self-evolutionary learning rate, call the improved Prophet model to output multi-step prediction results and prediction residuals within the validation time window, calculate the fitness value of each individual based on the fitness function and record the current best individual; Entering the iterative global search process, based on the bias jump direction adaptive mechanism, the search direction is determined according to the derivativeless difference evaluation result of the fitness function. When the threshold condition obtained from the predicted residual size and bias score is met, directional jump update is performed on the current individual, and the updated parameters are processed by feasible domain pruning and integerization of periodic coupling order parameters. In each iteration, based on the dynamic adjustment mechanism of abnormal perturbation amplitude, the confidence interval width of the improved Prophet model in multi-step prediction is used as an uncertainty measure. The perturbation step size is increased or decreased according to the uncertainty magnitude and the iteration decay factor is superimposed to keep the parameters falling within the preset boundary and maintain the legal values ​​of the periodic coupling order parameters. A population memory mechanism is used to maintain the elite solution set and the diverse solution set. The memory set is updated according to the fitness and inter-solution distance criteria. The search is terminated when the iteration limit is reached or the fitness convergence condition is met. The output is the optimal parameter set containing the temporal deformation factor parameter, the periodic coupling order parameter, and the dynamic self-evolutionary learning rate parameter.

7. The method for optimizing and controlling the quenching and finishing process of an internal combustion engine crankshaft according to claim 2, characterized in that, The process of retraining and predicting the standardized input sample set using the optimized Prophet model includes: The improved Prophet model is retrained by calling the obtained optimized parameters and setting the sliding training window length, prediction step size and confidence interval quantile level based on the standardized input sample set. After retraining, multi-step prediction is performed on quenching temperature, energy consumption, deformation, surface roughness and hardness. The predicted values ​​of each target, the corresponding prediction residuals and the upper and lower confidence intervals are output according to the prediction step size. Based on the obtained confidence interval width and recent forecast residual fluctuations, a risk-adjusted forecast sequence for control decisions is generated, and the future change trajectories of each objective are time-series organized. The sensitivity information of quenching current, scanning speed and cooling flow rate to each target is obtained from the retrained Prophet model. Combined with the target weights and reference target values, the control trend increment of quenching current, scanning speed and cooling flow rate in the prediction time domain is calculated. The obtained control trend increments are combined with the current process settings to form the dynamic change trends of quenching current, scanning speed and cooling flow rate.

8. The method for optimizing and controlling the quenching and finishing process of an internal combustion engine crankshaft according to claim 2, characterized in that, The calculation of the comprehensive cost function and determination of the optimal combination of control parameters includes: Based on the multi-step prediction results and corresponding confidence intervals, the set of optimization objectives is determined. Hardness dispersion, surface roughness prediction, energy consumption prediction and production cycle are taken as optimization objectives, and quenching current, scanning speed and cooling flow rate are taken as control parameters to be solved. The allowable value range and maximum variation of adjacent control times are set for quenching current, scanning speed and cooling flow rate, respectively. The length of the prediction time domain and the target weight configuration are set to form a decision space and evaluation criteria for solving the problem. Within the prediction time domain, the predicted values ​​of hardness dispersion, surface roughness, energy consumption, and production cycle time are weighted and summarized according to the target weights to construct a cost index for measuring comprehensive performance. The cost index reflects the overall balance between quality consistency, surface quality, energy efficiency, and cycle time. Combine the upper and lower limits of the confidence interval, set quality and geometric constraints to ensure that the lower limit of hardness is not lower than the preset threshold, the upper limit of surface roughness does not exceed the preset threshold, and the upper limit of deformation does not exceed the preset threshold, and ensure that the values ​​and variation ranges of quenching current, scanning speed and cooling flow rate meet the specified boundary and variation constraints. Under the premise of satisfying the constraints, the cost index is solved by multi-objective constraint to obtain the optimal combination of control parameters for quenching current, scanning speed and cooling flow rate.

9. The method for optimizing and controlling the quenching and finishing process of an internal combustion engine crankshaft according to claim 2, characterized in that, The process of applying the optimal control parameter combination to the quenching and finishing process of the internal combustion engine crankshaft, and performing parameter adjustment according to a time-rolling cycle, includes: Read the optimal combination of control parameters, generate a control execution plan that includes the target setpoints for quenching current, scanning speed and cooling flow rate, allowable value boundaries and the maximum variation range of adjacent cycles, and start parameter distribution according to the preset time rolling cycle; Within each rolling time cycle, the parameters of quenching current, scanning speed and cooling flow rate are adjusted sequentially according to the control execution plan, and the actual set values, effective time and adjustment results are recorded. During the parameter adjustment process, the measured results of quenching temperature, energy consumption, deformation, surface roughness and hardness, as well as the actual effective quenching current, scanning speed and cooling flow rate are collected simultaneously. Time alignment and data validity verification are completed to form execution result data containing the information required for prediction and comparison, and deviation information is generated. The execution results and deviation information are transmitted back to the multi-source process data acquisition stage in real time to update the time series deformation variables and periodic response quantities, trigger online retraining and parameter calibration of the improved Prophet model, and output the updated multi-step prediction results and confidence intervals as the basis for parameter adjustment and multi-objective constraint solution for the next time rolling cycle.

Images