Carburizing and quenching process parameter optimization method based on machine learning

Abstract

The invention discloses a carburizing and quenching process parameter optimization method based on machine learning, and relates to the technical field of carburizing and quenching process optimizing.The method comprises the steps that a multi-source sensor is arranged, environment and process data are collected in real time, and a high-quality synchronous data set is formed through unified timestamps and process labels; by means of the preprocessed data, an environment interference model is built through deep learning or time sequence regression, and the influence of external disturbance on the depth and hardness of a carburized layer is quantified; according to a model prediction result, a self-adaptive controller adjusts the carburizing agent flow, the heating power and the heat preservation duration in real time, closed-loop regulation and control are achieved, and it is ensured that the technological conditions and the product performance are always close to preset targets; the carburized layer and the hardness of the discharged part are detected, the actual result is fed back to the model for iterative updating, and the control strategy is continuously optimized, so that the problem of unstable quality caused by environmental fluctuation in the traditional process is effectively solved, the production efficiency is remarkably improved, and the energy consumption and the test cost are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of carburizing and quenching process optimization, and specifically to a method for optimizing carburizing and quenching process parameters based on machine learning. Background Art

[0002] In the fields of high-end equipment manufacturing and automotive transmission, gears, bearings, and various key components widely use carburizing and quenching processes to achieve both high hardness and good toughness, which plays a decisive role in the service life and reliability of the entire system. In recent years, with the continuous improvement of the automation level of industrial production lines, production workshops have gradually integrated online monitoring, intelligent scheduling and other functions, hoping to achieve real-time collection of multi-source environmental and process parameters such as temperature, humidity, air pressure, and carbon potential in the furnace, and then use data analysis methods for precise control and quality traceability. However, the actual production environment is often unable to maintain ideal conditions for a long time: the temperature and humidity of the factory will fluctuate greatly with the seasons and circadian rhythm; when the air pressure drops suddenly or the sealing of the equipment decreases over time, the carbon potential and furnace temperature distribution in the furnace will also become unstable; for small batches of parts with complex shapes and high surface quality requirements, flexible control strategies are needed to deal with uncertainties such as part stacking methods and differences in tooling heat dissipation.

[0003] Traditional "empirical" or "fuzzy interpolation" parameter adjustment methods typically rely solely on historically accumulated process curves, making small corrections within a static process window. This makes it difficult to keep up with the dynamic and ever-changing production pace and high-precision performance requirements. Furthermore, indicators such as carburized layer uniformity, hardness gradient, and quenching deformation exhibit nonlinear and strongly coupled relationships. Simply extending the holding time or increasing the heating power often leads to increased energy consumption and deformation risk, failing to meet the differentiated requirements of mass-produced and small-batch high-end parts.

[0004] However, the current process design and control still cannot completely get rid of the dependence on empirical or semi-empirical methods. The core problem lies in the lack of adaptive control means for the coupling effect of multi-source environmental data and process parameters. Specifically, when the temperature or humidity outside the workshop changes rapidly, the heat balance and carbon potential distribution inside the furnace seal will produce dynamic offsets, resulting in inconsistencies between the target carburized layer depth and the actual carburized layer formed; if the initial temperature of the tooling or the part itself is low and the surface heat absorption varies greatly, the same process curve will produce obvious hardness distribution deviations in different batches. With the accumulation of random external disturbances, in extreme cases, serious consequences such as insufficient carburizing, substandard hardness after quenching, or local overburning may occur, which not only increases the scrap rate but also affects the production capacity of the entire line; for high-value parts, the economic losses and delivery delays caused by such fluctuations are more obvious.

[0005] Traditional static PID control can only perform gentle corrections within a fixed range, which is not enough to quickly respond to process deviations under the action of multivariable coupling, and it is difficult to automatically identify and correct the impact of sudden environmental factors on the growth of the carburized layer and the surface structure transformation process; once the target and the actual result continue to deviate, manual intervention is required to perform temporary heating or delay operations, thereby increasing energy consumption and management costs. It can be seen that under the rapidly changing external environment and complex coupled process requirements, if there is a lack of an intelligent, dynamically learning and controllable process optimization system, it will be difficult to continuously maintain high-quality, consistent carburized layers and precise hardness distribution in actual production. This is the key technical problem that the present invention is committed to solving.

[0006] To this end, the present invention provides a method for optimizing carburizing and quenching process parameters based on machine learning. Summary of the Invention

[0007] (1) Technical problems solved

[0008] In response to the shortcomings of the existing technology, the present invention provides a method for optimizing carburizing and quenching process parameters based on machine learning. It collects environmental and process data in real time, and forms a high-quality synchronous data set with a unified timestamp and process label; uses the preprocessed data to construct an environmental interference model through deep learning or time series regression to quantify the impact of external disturbances on the depth and hardness of the carburizing layer; according to the model prediction results, the adaptive controller adjusts the carburizing agent flow, heating power and holding time in real time to achieve closed-loop control, ensuring that the process conditions and product performance are always close to the preset targets; by testing the carburizing layer and hardness of the furnaced parts, the actual results are fed back to the model for iterative updating, and the control strategy is continuously optimized, which significantly improves production efficiency, and reduces energy consumption and testing costs, thereby solving the technical problems recorded in the background technology.

[0009] (2) Technical solution

[0010] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0011] A machine learning-based method for optimizing carburizing and quenching process parameters involves invoking sensor arrays deployed in the workshop and furnace body to synchronously record operation timestamps and batch identifiers when a new batch of parts needs to be put into the furnace. After performing time series alignment and outlier removal on the collected data, a synchronously labeled dataset containing multi-dimensional environmental factors is generated.

[0012] By integrating the temperature field and carbon potential variation characteristics in the furnace, the key feature vectors that affect the depth and hardness of the carburized layer are extracted, and the environmental interference prediction value is output through the prediction function;

[0013] When there is a significant difference between the predicted environmental interference value and the preset target, the adaptive controller calculates the optimal process control parameters based on the comprehensive deviation function and adjusts the carburizing agent flow rate, heating power and holding time in real time to ensure carburizing uniformity and stable part performance.

[0014] After the parts are taken out of the furnace and the hardness and carburized layer depth tests are completed, the measured values are stored together with the optimal process control parameters and environmental interference prediction values in the output data set, and incremental training is performed on the environmental interference model and adaptive controller to gradually improve the process stability and accuracy in long-term and changing environments.

[0015] Preferably, a sensor array is arranged in each area of the carburizing and quenching workshop and inside the furnace body;

[0016] Each sensor periodically outputs a set of raw measurement values of multi-source environmental data, which are uniformly sent to the data acquisition module. The data are archived according to the sensor label k to form the original environmental data set D raw And carry out preliminary quality verification and preprocessing;

[0017] Preferably, in order to achieve time alignment of the outputs of each sensor, a unified reference clock t * , each piece of raw data is mapped to the corresponding reference clock to form a synchronized data record:

[0018] At key production nodes, operation labels L are assigned op And bind the corresponding time t op , associate the operation process with the environmental sensor data to obtain the synchronized labeling dataset D sync ;Introduction of Environmental Reference Index I env (t * ), used to simplify and summarize the overall level of multidimensional environmental parameters at a certain moment;

[0019] Preferably, from the synchronously marked dataset D sync , get the value with timestamp t * Multi-source environmental data, where each record carries its operation label L op ; Identify the data segments corresponding to different process stages of the same batch of parts based on the operation tags and organize them into groups;

[0020] Execute an abnormality elimination algorithm based on interval envelope or trend comparison on the time series readings of each sensor, mark the abnormal points and eliminate or mark compensation to obtain the stage data set of each process

[0021] Data sets at each stage of the process The entropy value is calculated and its variation within a certain time window is counted. If the entropy value is less than a certain threshold, it is considered to be information-poor and may need to be spliced with adjacent segments to ensure modeling availability. If the value is too large, it indicates that the sensor readings are changing rapidly and may require encrypted sampling or further aggregation processing.

[0022] Preferably, the corrected stage dataset As input, the sensor readings and operation labels L in the same batch or the same process op , combined with the environmental reference index I env (t * ), construct the environmental feature vector F(t * ):

[0023] F(t * )=[I env (t * ),X (1) (t * ),…,X (K) (t * ),L op (t * )]

[0024] The environmental feature vector F(t * ) according to the corresponding carburized layer depth, hardness test results or process deviation scalar (Δ c , Δ h etc.) for marking;

[0025] Using the environmental feature vector F(t * ) fits and outputs the predicted value of environmental interference Ω(t * ), and by minimizing the loss function The environmental interference model that can accurately map environmental characteristics to process errors or deviation trends is obtained, and the environmental feature vector F(t * ) corresponds to the predicted value of environmental interference Ω(t * )’s prediction output;

[0026] Preferably, the environmental interference prediction value Ω(t * ) to obtain the potential impact characterization of the current moment on the carburized layer depth and hardness; according to the batch identification and timestamp t * The predicted value of environmental interference Ω(t * ) and the target depth of the workpiece and target hardness Perform corresponding matching to form a triplet for subsequent control calculations

[0027] To introduce the comprehensive deviation value Measure the depth and hardness of the target and the degree of interference it may receive;

[0028] If the comprehensive deviation value is higher than the preset deviation threshold, strong adaptive control is started; if the comprehensive deviation value is higher than the preset deviation threshold, strong adaptive control is started; If the parameters are in a relatively safe range, only slight or regular adjustments are made to the parameters;

[0029] Preferably, the process variable vector u(t * ), combined with the comprehensive deviation value Constructing the control objective function

[0030] By minimizing the objective function Obtain process variable vector u(t * )’s optimal solution u * (t * ) so that the controller can issue real-time instructions to drive the carburizing agent valve, heater and holding timer to ensure that the process is adaptive to external interference;

[0031] In the sampling period, the adaptive controller is based on the latest batch of environmental disturbance prediction values Ω(t * ), and the comprehensive deviation value Quickly calculate the optimal solution u of the process variable vector * (t * ) and sent to the execution end;

[0032] When the comprehensive deviation value is significantly reduced or the objective function is adjusted When the value is effectively reduced, it means that the control strategy helps to reduce the influence of environmental interference on the depth and hardness of the carburized layer. If the comprehensive deviation value continues to rise, the control of u(t * ) correction amplitude to form a closed-loop control

[0033] Preferably, if during the control execution process, the output environmental disturbance prediction value Ω(t * ) or the on-site sensor suddenly detects extreme interference, causing the comprehensive deviation value to far exceed the difference deviation threshold Θ c , the adaptive controller immediately enters the abnormal compensation mode. If there is still no significant improvement, the reserved safety operation is forced to be enabled;

[0034] Define the safety limit parameter range, when the adaptive controller solves the optimal solution of the process variable vector u * (t * ) exceeds this range, it is truncated to the nearest end value;

[0035] Preferably, after the parts are taken out of the furnace, hardness testing and carburized layer depth measurement are performed to obtain the final performance indicators corresponding to the parts of this batch, and the test results are collected as the process variable vector u(t * ) and the corresponding environmental interference prediction value Ω(t * ), generate output data set D out ;

[0036] Preferably, the output dataset D is used out Evaluate the accuracy of the environmental interference model and adaptive controller in various scenarios and compare them with the measured depth With the goal The difference between the measured hardness and the With the goal The difference between the two, and combined with the environmental prediction Ω(t * ) output items for segmented deviation analysis;

[0037] Will add The samples are added to the training set or validation set of the environmental interference model, and the model parameters are retrained or fine-tuned;

[0038] Preferably, multiple batches of test data and corresponding environmental prediction and control parameters are regularly summarized to construct a long-term performance curve. If the long-term performance curve shows that certain auxiliary links repeatedly interfere with the carburizing quality, further optimization measures are introduced at the production planning level or equipment operation and maintenance level; the external parameters of the environmental interference model are periodically verified;

[0039] (3) Beneficial effects

[0040] The present invention provides a method for optimizing carburizing and quenching process parameters based on machine learning, which has the following beneficial effects:

[0041] Through multi-source environmental data collection and synchronous tagging, key sensor information such as temperature, humidity, air pressure, and carbon potential is effectively acquired and aligned, so that target indicators can be bound to the same time base, avoiding information distortion caused by traditional processes relying on single-point measurement. A unified batch identification and timestamp are also established to ensure that subsequent processing has a sufficient data foundation.

[0042] The constructed environmental interference model uses deep learning or time series regression methods to map the fused multi-dimensional features into the environmental interference prediction quantity Ω(t * ) and quantify the potential impact on carburized layer depth and hardness. This allows the nonlinear effects of factors such as temperature and humidity fluctuations and furnace door opening and closing to be captured before the process is executed, making the production process more forward-looking and explainable, and reducing deviations in hardness and depth indicators.

[0043] According to the environmental interference prediction quantity Ω(t *) and the target, and corrects core process parameters such as carburizing agent flow, heating power, and holding time in real time. When the interference amplitude is large, adaptive compensation and safety limit are triggered to avoid uneven depth layers and hardness overshoot caused by sudden environmental changes, significantly improving product consistency and yield rate.

[0044] By detecting measured values such as carburized layer depth and hardness and mapping them one-to-one with executed adaptive control data, the model can be retrained offline or online. This mechanism can continuously learn about equipment aging, sensor drift, and environmental changes under seasonal temperature differences, and iteratively optimize the interference model and control logic to ensure a balance between long-term production quality and energy consumption.

[0045] In summary, this solution perfectly combines environmental disturbance identification, adaptive parameter control, and continuous model iteration. It can not only significantly reduce product deviations caused by environmental fluctuations, but also continuously improve the system robustness and self-learning ability during long-term operation, providing support for the intelligent upgrade of carburizing and quenching processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The figure is a flow chart of the method for optimizing the carburizing and quenching process parameters of the present invention. DETAILED DESCRIPTION

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0048] See also Figure 1 The present invention provides a method for optimizing carburizing and quenching process parameters based on machine learning, comprising:

[0049] Step 1: When a new batch of parts is detected and needs to be put into the furnace, the sensor arrays deployed in the workshop and the furnace body are called to synchronously record the operation timestamp and batch identification. After performing time series alignment and outlier removal on the collected data, a synchronously labeled dataset containing multi-dimensional environmental factors is generated;

[0050] The step 1 includes the following:

[0051] Step 101: Multi-source environmental data sensor arrangement and raw information acquisition

[0052] A sensor array is arranged in each area of the carburizing and quenching workshop and inside the furnace, including several temperature sensors T (k) 、Humidity sensor H (k) , air pressure sensor P (k) and carbon potential sensor C(k) , where k is the sensor number;

[0053] Each sensor periodically outputs a set of raw measurement values of multi-source environmental data, denoted as T (k) 、H (k) (t), P (k) (t) and C (k) (t), t is the sampling time, the above raw measurement values are sent to the data acquisition module, and the data acquisition module archives the data according to the sensor label k to form the original environmental data set D raw And perform preliminary quality verification and preprocessing. If the sensor is offline or the reading exceeds the limit threshold, it will be marked as an abnormal point in the data structure;

[0054] Step 102: Timestamp synchronization and environmental reference value marking

[0055] In order to achieve the time alignment of each sensor output, a unified reference clock t * , each piece of original data T (k) (t), H (k) (t), P (k) (t), C (k) (t) are mapped to the corresponding reference clock to form a synchronous data record:

[0056] [t * ,T (k) (t * ),H (k) (t * ),P (k) (t * ),C (k) (t * )]

[0057] At key production nodes such as parts entering the furnace, heating, carburizing, and quenching, the operation label L is assigned respectively. op And bind the corresponding time t op , associate the specific operation process with the environmental sensor data to obtain the synchronous labeling dataset D sync ;

[0058] Introduction of Environmental Reference Index I env (t * ), which is used to simplify and summarize the overall level of multidimensional environmental parameters at a certain moment, is defined as follows:

[0059]

[0060] Where: X(u) is the multi-source sensor reading vector collected at time u, defined as:

[0061]

[0062] Among them, X (k) (u) represents the measurement value of the kth sensor (such as temperature, humidity, air pressure or carbon potential) at time u, and K is the total number of sensors; Φ(X(u)) is the nonlinear conversion function vector for X(u), which is used to crop, amplify or map different measurement dimensions or ranges, and is defined as:

[0063]

[0064] Φ k (·) is a single sensor mapping function; it performs nonlinear mapping on the sensor readings to eliminate the imbalanced value range caused by different dimensions or different proportions. It can generally be set as a monotonically increasing function, a piecewise function, or an interval value processing function;

[0065] W is a diagonal weight matrix of size K×K:

[0066] W=diag[w1,w2,…,w K ]

[0067] where w k is the weight coefficient of the kth sensor in the environmental reference index, which is used to adjust the influence of different sensor quantities on the overall index; ||·|| p L represents the vector p norm, p is the spatial norm index, specifically:

[0068]

[0069] The corresponding vector WΦ(X(u)) is:

[0070]

[0071] γ is the additional norm exponent, the value range is: γ>1; Δt is the value at the current time t * The size of the time window for looking back;

[0072] Environmental Reference Index I env (t * ) can be calculated continuously in the time dimension to measure the magnitude of environmental changes within a time series, providing a more directional numerical basis for subsequent modeling and real-time regulation;

[0073] When used, calculate the environmental reference index I env (t * ) can extract a single index from high-dimensional sensor data to quickly judge the overall state of the environment, and the key operation label L opBinding with timestamps allows subsequent environmental interference models to not only identify the discrete degree of environmental parameters but also directly trace their specific impact locations in each step of the process, strengthening the correlation analysis between process and environment. It can also provide high-quality input with operating scenarios and comprehensive indexes, achieving a smooth transition from multi-source sensing to model training.

[0074] Step 2: Integrate the temperature field and carbon potential change characteristics in the furnace to extract the key feature vectors that affect the depth and hardness of the carburized layer, and output the environmental interference prediction value through the prediction function;

[0075] The second step includes the following:

[0076] Step 201: Data consistency processing and noise anomaly identification

[0077] From the synchronously labeled dataset D sync , get the value with timestamp t * Multi-source environmental data (temperature, humidity, air pressure, carbon potential, etc.), where each record carries its operation label L op ; Identify the data segments corresponding to the same batch of parts at different process stages based on the operation tags and group them into

[0078] An abnormality rejection algorithm based on interval envelope or trend comparison is performed on the time series readings of each sensor, where:

[0079]

[0080] Among them, X (k) (t * ) is the kth sensor at time t * The value of μ k (t * ) is the local smoothing center (which can be obtained by history window or regression method), ω k is an adjustable amplification factor, the value is greater than 0, Υ(t * ) is a customizable envelope or trend function, for example, the exponentially weighted moving average (EWMA) function can be used; if ψ = 1, it is marked as an outlier and removed from subsequent analysis or marked for compensation. After consistency alignment, noise filtering, and outlier removal, the stage data set of each process is obtained.

[0081] Data sets at each stage of the process Within the data, information entropy methods (discrete Shannon entropy formula) or complexity measures based on interval distance are introduced to count the amplitude of its change within a certain time window. If the entropy value is less than a certain threshold, it is considered to be information-poor and may need to be spliced with adjacent segments to ensure modeling availability. If the value is too large, it indicates that the sensor readings are changing rapidly and may require encrypted sampling or further aggregation processing.

[0082] This system ensures the consistency of environmental data across time and operational tags. It also eliminates suspicious readings or low-quality fragments through targeted anomaly identification and periodic data quality assessment, laying a highly reliable data foundation for building environmental interference models. It also enables flexible, multi-scale anomaly detection; combining information entropy or interval distance measurement allows it to adapt to varying levels of fluctuation at different process stages, demonstrating high scalability.

[0083] Step 202: Multi-dimensional construction and prediction output of environmental interference model

[0084] With the corrected stage data set As input, for the sensor readings in the same batch or process {X (1) (t * ), X (2) (t * )…,X (K) (t * )} and operation label L op , combined with the environmental reference index I env (t * ), construct the environmental feature vector F(t * ):

[0085] F(t * )=[I env (t * ),X (1) (t * ),…,X (K) (t * ),L op (t * )]

[0086] The environmental feature vector F(t * ) according to the corresponding carburized layer depth, hardness test results or process deviation scalar (Δ c , Δ h etc.) to provide supervision targets for subsequent environmental interference model training;

[0087] Construct environmental interference models through deep neural networks (such as time series convolution models) or regression based on exponential kernel functions;

[0088] Using the environmental feature vector F(t * ) fits and outputs the predicted value of environmental interference Ω(t * ), environmental interference prediction value Ω(t * ) aims to quantify the * The comprehensive deviation trend of the environment on the depth and hardness of the carburized layer when

[0089] In order to reflect the strengthening of nonlinearity and extreme value effects, the following formula can be used as the loss function

[0090]

[0091] Where: y(u) represents the actual measured carburized layer depth and hardness index (or process deviation) in the time interval The truth value sequence within ; is the corresponding sequence of the model prediction output; Γ is a predefined or adjustable nonlinear mapping used to highlight large errors or errors in a specific interval; η is the power exponent that adjusts the sensitivity of the integral to the overall loss, and its value is greater than 0;

[0092] By minimizing the loss function An environmental interference model that can accurately map environmental characteristics to process errors or deviation trends can be obtained, and the environmental feature vector F(t * ) corresponds to the predicted value of environmental interference Ω(t * )’s predicted output.

[0093] The output of the environmental interference model is the environmental interference prediction value Ω(t * ) to perform buffer smoothing or time series extrapolation to obtain the environmental interference prediction result, and the environmental interference prediction value Ω(t * ) Regular preservation can help with traceability and comparison during continuous process iteration and update, so as to continuously optimize the model in the long term;

[0094] When used, advanced modeling techniques such as deep learning or kernel regression are used to capture complex environment-process relationships in multidimensional feature space and to predict the process through nonlinear loss. Enhanced fitting capabilities for extreme deviations enable the model to more accurately quantify the effects of environmental disturbances on carburizing depth and hardness;

[0095] By predicting the environmental interference value Ω(t * ) can perform time series extrapolation to predict the process deviation risks caused by environmental fluctuations in advance and provide highly timely decision-making information for adaptive parameter adjustment.

[0096] Step 3: When there is a significant difference between the predicted environmental interference value and the preset target, the adaptive controller calculates the optimal process control parameters based on the comprehensive deviation function and adjusts the carburizing agent flow rate, heating power and holding time in real time to ensure carburizing uniformity and stable part performance;

[0097] The step three includes the following:

[0098] Step 301: Multi-index quantification of target and interference mapping

[0099] The environmental interference prediction value Ω(t * ) to obtain the potential impact characterization of the current moment on the carburized layer depth and hardness;

[0100] According to the batch identification and timestamp t * The predicted value of environmental interference Ω(t * ) and the target depth of the workpiece and target hardness Perform corresponding matching to form a triplet for subsequent control calculations

[0101] In order to simultaneously measure the degree of interference that the depth and hardness targets may receive, a comprehensive deviation value is introduced Defined as:

[0102]

[0103] in: Used to calculate the degree of deviation of carburized layer depth caused by environmental interference; Used to calculate the degree of hardness deviation caused by environmental interference; β is a power exponent that can amplify the penalty for large deviations and its value is greater than 0;

[0104] in Using the following formula, Similarly:

[0105]

[0106] Where: Ω c (t * ) indicates that at time t * Carburized layer depth predicted by the environmental interference model; Indicates the target carburized layer depth required by the process; τ c It is the depth tolerance parameter, which is used to set the allowable error within the normal process fluctuation range. When the actual deviation is lower than τ c When , the function output is zero, indicating that the deviation is within the allowable range;

[0107] If the comprehensive deviation value is higher than the preset deviation threshold, it means that under the established parameter settings, environmental disturbances may cause a large deviation in the carburized layer depth or hardness, and strong adaptive control needs to be started; if If the value is in a relatively safe range, only slight or regular adjustments are made to the parameters;

[0108] When used, the environmental interference prediction Ω(t * ) is closely mapped with the process target (depth and hardness) to generate a numerical comprehensive deviation value Provide a clear and adjustable quantitative standard for the trigger mechanism to be adjusted in real time.

[0109] Step 302: Parameter update strategy and real-time control implementation

[0110] Define the process variable vector u(t * ):

[0111] u(t * )=[u C (t * ),u P (t * ),u T (t * )]

[0112] Where: u C (t * ) represents the real-time carburizing agent flow coefficient, which is used to dynamically adjust the supply of carburizing agent during the carburizing process; u P (t * ) represents the heating power correction amount, which represents the correction amplitude of the heater output power in real-time process control; u T (t * ) represents the insulation time offset, which is used to adjust the actual maintenance time of the insulation stage, offset relative to the preset insulation time; combined with the comprehensive deviation value Constructing the control objective function as follows:

[0113]

[0114] Among them, Γ(·) is a nonlinear function, which takes the comprehensive deviation value Converted into contribution to control accuracy;

[0115] γ(·) is used to characterize the energy consumption or additional cost of the adjustment variable. For example, excessively increasing the heating power will increase energy consumption, which also needs to be comprehensively balanced during regulation. α1 and α2 are weight coefficients, with values between 0 and 1.

[0116] By minimizing the objective function Obtain process variable vector u(t * )’s optimal solution u * (t * ) so that the controller can issue real-time instructions to drive the carburizing agent valve, heater and holding timer to ensure that the process is adaptive to external interference;

[0117] In the sampling period, the adaptive controller is based on the latest batch of environmental disturbance prediction values Ω(t * ) and comprehensive deviation value Quickly calculate the optimal solution u of the process variable vector * (t * ) and sent to the execution end;

[0118] When the comprehensive deviation value is significantly reduced or the objective function is adjusted When the value is effectively reduced, it means that the control strategy helps to reduce the influence of environmental interference on the depth and hardness of the carburized layer. If the comprehensive deviation value continues to rise, the control of u(t * ) correction amplitude to form a closed-loop control

[0119] When used, a control target is established that can integrate multiple factors (process quality requirements, energy consumption, disturbance prediction), and based on The size of the process variable vector u(t * ) vector schedules multiple key parameters at one time, which is more conducive to maintaining carburizing uniformity under complex disturbance conditions.

[0120] Step 303: Abnormal feedback compensation and dynamic safety limit

[0121] If during the control execution process, the output environmental disturbance prediction value Ω(t * ) or the on-site sensor suddenly detects extreme interference (such as a sudden drop in air pressure, abnormal tooling temperature), causing the comprehensive deviation value to far exceed the difference deviation threshold Θ c , the adaptive controller immediately enters the abnormal compensation mode;

[0122] In the abnormal compensation mode, the weight coefficient α1 is temporarily increased to prioritize reducing the deviation caused by the environment. If there is still no significant improvement, the reserved safety operations are forcibly activated, such as quickly shortening the holding time and reducing the heating power to avoid extreme situations such as overheating of the workpiece or supersaturated carburization;

[0123] As a further step, in order to prevent frequent and large-scale parameter adjustments from causing violent fluctuations in furnace temperature or carbon potential, a set of safety limit parameters u is defined. min with u max ,satisfy:

[0124] u min ≤u(t * )≤u max

[0125] Where: u min and u max are the minimum and maximum operating boundaries set based on the physical performance of the equipment and the process limit requirements respectively; when the adaptive controller solves the optimal solution of the process variable vector u * (t * ) exceeds this range, it will be truncated at the nearest end value to ensure production line safety and equipment life;

[0126] The process variable vector u(t * ) and the corresponding comprehensive deviation value The change curve is recorded together.

[0127] During use, on the basis of normal real-time optimization, abnormal compensation and limiting mechanisms are introduced to ensure that the system will not lose control due to extreme disturbances or short-term sensor deviations, thereby improving process safety and stability; by recording the compensation process under abnormal conditions, the model can be gradually trained in subsequent iterative updates on how to more effectively deal with rare but huge interference scenarios, combining environmental disturbances with process goals, using a nonlinear method to quickly determine the optimal parameter adjustment scheme, and comprehensively using compensation mechanisms and safety limits to take into account both flexibility in dealing with extreme disturbances and process safety.

[0128] Step 4: After the parts are released from the furnace and the hardness and carburized layer depth are tested, the measured values are stored together with the optimal process control parameters and environmental interference prediction values in the output data set. Incremental training is then performed on the environmental interference model and adaptive controller to gradually improve process stability and accuracy in long-term and changing environments.

[0129] The step 4 includes the following contents:

[0130] Step 401: Finished product testing and data integration

[0131] After the parts are fired, hardness test (such as Vickers or Rockwell hardness test) and carburized layer depth test (such as metallographic analysis or automatic depth measurement) are performed to obtain the final performance index of the parts in this batch. Form record, where Indicates the measured carburized layer depth, Indicates the measured hardness value;

[0132] Collect the process variable vector u(t * ) and the corresponding environmental interference prediction value Ω(t *); and ensure that it is consistent with and By batch identification L batch or timestamp t * ——Corresponding; generate output data set D out , its basic data elements can be described as follows:

[0133]

[0134] Facilitates accurate traceability and comparison during subsequent model training; actual finished product performance of this batch The environmental parameters (Ω(t * )) and process control u(t*) are strongly correlated to form a complete data closed loop, ensuring that subsequent model evaluation and iterative training are more targeted.

[0135] Step 402: Model evaluation and online / offline retraining

[0136] Using the output dataset D out Evaluate the accuracy of the environmental interference model and the adaptive controller in various scenarios. The adaptive controller is used to calculate and adjust key parameters of the carburizing and quenching process in real time, which can be obtained through machine learning algorithm training:

[0137] Compare to measured depth With the goal The difference between the measured hardness and the With the goal The difference between the two, and combined with the environmental prediction Ω(t * ) output term (such as Ω c (t * ),Ω h (t*)) to conduct segmental deviation analysis;

[0138] Will add The samples are added to the training set or validation set of the environmental interference model, and the model parameters are retrained or fine-tuned offline (such as batch update during non-production period) or online (small batch update during production). To prevent the control strategy from being unstable due to frequent iteration of the model, the adaptive rate κ can be set. a , triggering the official version update after accumulating a certain amount of data in multiple batches;

[0139] When used, by incorporating newly added finished product inspection results into the training process, the environmental interference model can continuously adapt to seasonal or extreme environmental changes, avoiding the gradual decline in model accuracy due to environmental conditions or part batch changes. At the same time, the adaptive controller can continuously evolve to cope with error accumulation or special scenarios that occur under real production conditions, ensuring long-term steady-state performance. Flexibly combine offline retraining with online small-batch updates and set the adaptive rate κa Together with version management, it ensures timely error correction without introducing multiple overfitting or control oscillations due to frequent updates. Special parts from different batches or extreme environments can be given higher weights in the process, forming a hierarchical retraining mechanism with more refined adaptability.

[0140] Step 403: Long-term performance recording and global optimization feedback

[0141] Regularly summarize multiple batches of test data and corresponding environmental prediction and control parameters, and evaluate the overall stability and energy consumption level of the carburizing and quenching process under different periods and environmental fluctuations by time series analysis or period comparison;

[0142] On this basis, long-term performance curves can be constructed to globally observe the evolution trend of carburized layer depth and hardness pass rate over time and identify potential degradation (such as equipment aging, sensor inaccuracy) or breakthrough points (process upgrade);

[0143] If long-term performance analysis shows that certain auxiliary links (such as cooling oil temperature maintenance and furnace door sealing performance) repeatedly interfere with carburizing quality, further optimization measures (such as replacing sealing components and correcting cooling processes) can be introduced at the production planning level (MES system) or equipment operation and maintenance level to reduce frequent interference with core carburizing control;

[0144] Periodically check the external parameters of the environmental interference model (such as the air pressure detection channel and the furnace gas supply system) to prevent the accumulation of model errors due to hardware aging or sensor offset;

[0145] On the workshop big data platform or MES system, the accumulated time series analysis results are compared with economic indicators (such as energy consumption, output, and scrap rate) in multiple dimensions to facilitate making better production scheduling or process configuration decisions based on real historical data;

[0146] When in use, single or short-term process results are expanded to long-term trend monitoring and global decision-making optimization, and local improvements of independent batches are gradually transformed into systematic improvements for the entire production line and the entire cycle; through cumulative records and time series analysis, insights into the aging patterns of equipment and sensors can be obtained, hidden dangers can be discovered earlier and repaired, ensuring continuous and reliable operation of the process.

[0147] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0148] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0149] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is only for some logical functions. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0150] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0151] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for optimizing carburizing and quenching process parameters based on machine learning, characterized by: include, When a new batch of parts is detected entering the furnace, the sensor arrays deployed in the workshop and the furnace are called upon to perform time series alignment and outlier removal on the acquired data to generate a synchronized labeled data set containing multi-dimensional environmental factors. By integrating the temperature field and carbon potential variation characteristics in the furnace, the key feature vectors that affect the depth and hardness of the carburized layer are extracted, and the environmental interference prediction value is output through the prediction function; When there is a significant difference between the environmental interference prediction value and the preset carburizing layer target and hardness target, the adaptive controller calculates the optimal process control parameters based on the comprehensive deviation value and adjusts the carburizing agent flow rate, heating power and holding time in real time; After the parts are taken out of the furnace and the hardness and carburized layer depth are tested, the measured values, optimal process control parameters and environmental interference prediction values are stored in the output data set, and incremental training is performed on the environmental interference model and adaptive controller.

2. The method for optimizing carburizing and quenching process parameters based on machine learning according to claim 1, wherein: Each sensor periodically outputs a set of raw measurement values of multi-source environmental data, which are sent to the data acquisition module. The data is archived according to the sensor label to form a raw environmental data set and undergoes preliminary quality verification and preprocessing. Operation labels are assigned to key production nodes and bound to corresponding times. The operating procedures are associated with environmental sensor data to obtain a synchronized labeled dataset; and an environmental reference index is introduced to simplify and summarize the overall level of multidimensional environmental parameters at a certain moment.

3. The method for optimizing carburizing and quenching process parameters based on machine learning according to claim 2, wherein: Acquire multi-source environmental data from the synchronously labeled dataset, identify the data segments corresponding to different process stages of the same batch of parts based on the operation labels, and organize them into groups; The abnormal rejection algorithm is executed on the time series readings of each sensor, marking the abnormal points and eliminating or marking compensation, obtaining the stage data set of each process, and counting its change range within a certain time window.

4. The method for optimizing carburizing and quenching process parameters based on machine learning according to claim 3, characterized in that: Taking the corrected stage dataset as input, the environmental feature vector is constructed in the multidimensional space by combining sensor readings and operation labels with environmental reference indices. The environmental characteristic vectors in different time periods are marked according to their corresponding carburized layer depth, hardness test results or process deviation scalars; The environmental feature vector is used to fit and output the environmental interference prediction value, and the environmental feature vector at each moment is matched to the prediction output of the environmental interference prediction value.

5. The method for optimizing carburizing and quenching process parameters based on machine learning according to claim 4, characterized in that: The environmental interference prediction value is matched with the target depth and target hardness of the workpiece according to the batch identification and timestamp to form a triplet for subsequent control calculations. A comprehensive deviation value is introduced to measure the degree of interference to the depth and hardness targets. If the comprehensive deviation value is higher than the preset deviation threshold, strong adaptive control is started. If the comprehensive deviation value is in a relatively safe range, only slight or regular adjustments are made to the parameters.

6. The method for optimizing carburizing and quenching process parameters based on machine learning according to claim 5, characterized in that: Define the process variable vector that can be adjusted by the adaptive controller, and construct the control objective function based on the comprehensive deviation value; By minimizing the control objective function, the optimal solution of the process variable vector is obtained, and the controller issues real-time instructions to drive the carburizing agent valve, heater, and holding timer, ensuring that the process is adaptive to external interference. When the comprehensive deviation value does not decrease significantly or the control objective function is not effectively reduced, the correction amplitude of the process variable vector is increased in the next sampling period.

7. The method for optimizing carburizing and quenching process parameters based on machine learning according to claim 6, characterized in that: If, during control execution, the output environmental interference prediction value or the on-site sensor suddenly detects extreme interference, causing the comprehensive deviation value to far exceed the difference deviation threshold, the adaptive controller will immediately enter the abnormal compensation mode; if there is still no significant improvement, the reserved safety operation will be forcibly enabled; A safety limit parameter interval is defined. When the optimal solution of the process variable vector calculated by the adaptive controller exceeds the interval, it is truncated at the nearest end value.

8. The method for optimizing carburizing and quenching process parameters based on machine learning according to claim 7, characterized in that: After the parts are fired, hardness test and carburized layer depth measurement are performed to obtain the corresponding performance indicators of this batch of parts; The process variable vectors as adaptive control parameters and the corresponding environmental interference prediction values are collected to generate an output data set.

9. The method for optimizing carburizing and quenching process parameters based on machine learning according to claim 8, characterized in that: Use output datasets to evaluate the accuracy of environmental disturbance models and adaptive controllers in a variety of scenarios; Compare the difference between the measured data and the target, and perform segmented deviation analysis based on the output of the environmental prediction. Add the newly added samples to the training set or validation set of the environmental interference model, and retrain or fine-tune the model parameters.

10. The method for optimizing carburizing and quenching process parameters based on machine learning according to claim 9, characterized in that: Regularly summarize multiple batches of test data and corresponding environmental prediction and control parameters to build long-term performance curves; If the long-term performance curve shows that certain auxiliary links repeatedly and significantly interfere with the carburizing quality, further optimization measures should be introduced at the production planning level or equipment operation and maintenance level, and the external parameters of the environmental interference model should be periodically checked.

Images