Forging quenching and tempering method based on intelligent temperature control

Abstract

The invention relates to the technical field of forging heat treatment, and discloses a forging quenching and tempering method based on intelligent temperature control. The method comprises the following steps: acquiring a real-time temperature time sequence and a temperature distribution characteristic during heat treatment of a forge piece, and executing multi-scale signal decomposition through an optimal signal decomposition primary function and a decomposition depth hierarchy to obtain a multi-scale temperature signal component; and calculating a complexity metric value of the decomposition coefficient to extract randomness characteristics, and screening out a key temperature region coefficient which is most relevant to the structure evolution of the forge piece. The optimal temperature control frequency parameter is determined by combining the material phase change baseline characteristic, the current heat treatment stage state and the randomness characteristic, a temperature field evolution characteristic sequence is reconstructed and is modally decomposed into a slow change component and an accelerated change component, accordingly, the hardening and tempering state type is recognized, a temperature control strategy library is matched to generate a control instruction, and accurate and efficient forging hardening and tempering treatment is achieved. The method solves the problem of lagging structure performance regulation in traditional heat treatment, and improves the stability and consistency of the comprehensive performance of the forge piece.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of forging heat treatment, in particular to a forging quenching and tempering method based on intelligent temperature control. BACKGROUND

[0002] Forging quenching and tempering is a key process that combines quenching and high-temperature tempering to make forgings have excellent comprehensive mechanical properties, and is widely used in mechanical manufacturing, aerospace, automobile industry and other fields. In the traditional forging quenching and tempering process, temperature control mainly depends on empirical parameter setting or simple closed-loop feedback adjustment, which is difficult to adapt to complex and changeable heat treatment environment.

[0003] The existing temperature control method usually takes the temperature data of a single temperature measurement point as the basis for regulation and control, ignoring the spatial difference of temperature distribution inside and on the surface of the forging. Due to the complex shape and various sizes of forgings, the heat conduction rate and heat capacity of different regions differ significantly, and the temperature value of a single temperature measurement point cannot fully reflect the real state of the overall temperature field, which may easily lead to local temperature being too high or too low, thereby affecting the uniformity of phase change.

[0004] The traditional method mainly stays at the level of simple filtering or smoothing in processing the temperature time series, and fails to deeply mine the multi-scale features contained in the temperature signal. Temperature changes show different rules at different time scales. Fluctuations at short time scales may reflect transient interference of equipment operation, while trends at long time scales are related to the overall process. The existing technology cannot effectively distinguish these features, resulting in lagging or excessive temperature control response.

[0005] The phase change behavior of forging materials is closely related to temperature changes, and the phase change dynamics of different materials at different heat treatment stages differ significantly. The existing temperature control strategy lacks dynamic correlation with the baseline features of material phase change, and it is difficult to adjust the control parameters according to the real-time phase change state, often resulting in insufficient or excessive phase change, affecting the final microstructure and performance of forgings.

[0006] In terms of state recognition, the traditional method mainly relies on manual experience or offline detection means to judge the quenching and tempering state, and cannot capture the dynamic process of microstructure evolution in real time. This lag makes it difficult to adjust the process in real time, resulting in poor consistency of forging quality and high scrap rate in batch production. These problems jointly restrict the precision and stability of forging quenching and tempering, and it is difficult to meet the stringent requirements of high-end manufacturing on forging performance. SUMMARY

[0007] The present application aims to provide a forging quenching and tempering method based on intelligent temperature control to solve the problems raised in the background.

[0008] To achieve the above objectives, the present invention provides a method for tempering forgings based on intelligent temperature control, the method comprising: The real-time temperature time series and corresponding temperature distribution characteristics of the forging during the heat treatment process are obtained, and the optimal signal decomposition basis function and decomposition depth level corresponding to the temperature distribution characteristics are obtained. These are used to perform multi-scale signal decomposition on the real-time temperature time series to obtain multi-scale temperature signal components. Obtain the complexity metric of the decomposition coefficients corresponding to the multi-scale temperature signal components in order to extract the randomness characteristics of the decomposition coefficients; Based on the multi-scale temperature signal components and randomness characteristics, the key temperature region coefficients most relevant to the evolution of forging microstructure are selected from multiple decomposition coefficients. The baseline characteristics of material phase transformation, the current heat treatment stage state, and the randomness characteristics of the forging are obtained to determine the optimal temperature control frequency parameters that are suitable for the material properties. Based on the key temperature region coefficient and the optimal temperature control frequency parameter, a temperature field evolution characteristic sequence is reconstructed; mode decomposition is performed on the temperature field evolution characteristic sequence to obtain the slow-change component reflecting the steady-state evolution of the tissue and the rapid-change component reflecting the dynamic response of the phase transition. Based on the slow-change components of the steady-state evolution of the structure and the rapid-change components of the phase transformation dynamic response, the corresponding tempering state type of the forging is identified. According to the tempering state type, a preset temperature control strategy library is matched to generate control instructions for the heat treatment equipment to achieve tempering of the forging.

[0009] Preferably, obtaining the optimal signal decomposition basis function and decomposition depth level corresponding to the temperature distribution characteristics includes: Select the functional characteristic parameters of the candidate basis functions, wherein the functional characteristic parameters include at least the support interval length and the regularization order; By inputting the function characteristic parameters, the temperature distribution features, and the preset decomposition depth range through a swarm intelligence algorithm, the optimal signal decomposition basis function and its corresponding decomposition depth level are determined from the candidate basis functions.

[0010] Preferably, the step of performing multi-scale signal decomposition on the real-time temperature time series to obtain multi-scale temperature signal components includes: The real-time temperature time series is processed using a fast decomposition algorithm based on the optimal signal decomposition basis function and decomposition depth level; edge effects are suppressed at the sequence boundaries using a mirror continuation method. Extract a set of decomposition coefficients at multiple scales, and construct the multi-scale temperature signal components based on the set of decomposition coefficients.

[0011] Preferably, the acquisition of the material phase transformation baseline characteristics, current heat treatment stage state, and randomness characteristics of the forging includes: The thermal response characteristics of materials under different phase transition states were obtained through multi-stage heat treatment experiments, wherein the multi-stage process includes at least a heating stage, a holding stage, and a cooling stage. Based on the material's thermal response characteristics, the current heat treatment stage state and randomness characteristics are extracted.

[0012] Preferably, determining the optimal temperature control frequency parameter to suit the material properties includes: Based on the frequency domain energy distribution characteristics, current heat treatment stage state, and randomness characteristics of the material phase transition baseline features, a multi-layer fusion scoring calculation is performed on the preset candidate frequency set. The scoring includes phase transition matching degree score, stage state fitness score, and randomness correlation score. The weighting coefficients of the score are dynamically adjusted using an optimization algorithm to determine the optimal temperature control frequency parameter from the candidate frequency set.

[0013] Preferably, the step of reconstructing and generating a temperature field evolution feature sequence based on the key temperature region coefficient and the optimal temperature control frequency parameter includes: A spatiotemporal mapping matrix is ​​constructed, wherein the matrix elements of the spatiotemporal mapping matrix are generated by nonlinear combination of the key temperature region coefficient, the frequency matching degree corresponding to the optimal temperature control frequency parameter, and the preset phase change modulation coefficient; Multi-scale time series analysis is performed on the spatiotemporal mapping matrix. The multi-scale time series analysis includes short-time analysis to capture phase transition characteristics, medium-time analysis to track the thermal field stability state, and long-time analysis to monitor tissue evolution trends. By adaptively weighting and fusing the results of the multi-scale time series analysis, a temperature field evolution feature sequence containing multimodal phase transition characteristics is generated.

[0014] Preferably, performing mode decomposition on the temperature field evolution characteristic sequence to obtain the slow-change component reflecting the steady-state evolution of the tissue and the rapid-change component reflecting the dynamic response of the phase transition includes: The components of the temperature field evolution feature sequence are subjected to extreme point detection and boundary extension processing, and multi-mode components that satisfy the intrinsic conditions are obtained through iterative screening. Based on the average period and energy distribution characteristics of the multi-mode components, the earlier short-period components are classified as rapid-change components, and the later long-period components and residual terms are classified as slow-change components.

[0015] Preferably, the identification of the quenching and tempering state type corresponding to the forging includes: Extract the time-domain statistical features, frequency band energy distribution features, and inter-component correlation features of the gradual variation components to construct a steady-state feature vector; The transient amplitude change rate, peak characteristics and waveform structure features of the rapid change component are extracted to construct a transition feature vector; A hierarchical classification strategy is used to identify the state of the steady-state feature vector, and a threshold detector is used to identify the phase transition event corresponding to the transition feature vector. Based on the fusion of the state recognition results and phase transition events using a probabilistic network, and combined with spatiotemporal continuity features, confidence assessment is performed to output the heat treatment state type.

[0016] Preferably, the step of matching a preset temperature control strategy library to generate control commands for the heat treatment equipment includes: The heat treatment state type is mapped and matched with the temperature curve template, heat preservation time parameter and cooling rate constraint in the strategy library; The control command, which includes the target temperature value, the heating slope, and the stage switching conditions, is generated based on the matching results.

[0017] Preferably, after generating the control command for the heat treatment equipment, it further includes: Real-time monitoring of surface temperature distribution and internal heat flow characteristics of forgings, and calculation of the deviation metric between the current temperature field and the target temperature curve; When the deviation metric exceeds a preset threshold, the real-time temperature time series is reacquired and the temperature distribution characteristics are updated to iteratively adjust the control commands.

[0018] Compared with the prior art, the beneficial effects of the present invention are: This intelligent temperature control-based forging tempering method achieves refined management of the forging tempering process through multi-dimensional temperature signal analysis and dynamic control strategies. In the temperature signal processing stage, by acquiring real-time temperature time series and corresponding temperature distribution characteristics, and combining the optimal signal decomposition basis function and decomposition depth level for multi-scale signal decomposition, it can comprehensively capture the temperature change patterns at different time scales. This encompasses both macroscopic temperature features reflecting the overall process trend and microscopic temperature information reflecting local subtle fluctuations, avoiding the omission of key temperature features by traditional single-scale analysis.

[0019] By extracting the complexity metric of the decomposition coefficients to obtain randomness characteristics, a quantitative basis is provided for subsequent selection of key coefficients. These random characteristics can effectively reflect the fluctuation patterns in temperature signals, distinguish between irregular fluctuations caused by equipment interference and regular changes caused by microstructure evolution, so that the selected key temperature region coefficients can more accurately correlate with the microstructure evolution process of forgings and reduce the interference of irrelevant signals on control decisions.

[0020] In determining the temperature control parameters, by combining the baseline characteristics of the forging's material phase transformation, the current state of the heat treatment stage, and its randomness, the temperature control frequency parameters can be adapted to the phase transformation characteristics of the material at different heat treatment stages. The sensitivity to temperature changes varies during the material's phase transformation process; dynamically adjusted temperature control frequencies can enhance control sensitivity during the active phase transformation stage and maintain control stability during the steady-state stage, avoiding the problems of insufficient response or over-adjustment during the critical phase transformation period in traditional fixed-parameter control.

[0021] Based on the temperature field evolution feature sequence reconstructed using key temperature region coefficients and optimal temperature control frequency parameters, the gradual change component and rapid change component obtained through modal decomposition correspond to the steady-state evolution of the microstructure and the dynamic response process of phase transformation, respectively. This detailed subdivision makes the identification of the microstructure evolution state of forgings more accurate, clearly capturing key nodes such as the onset, activity, and end of phase transformation, and providing multi-dimensional basis for the identification of quenching and tempering state types.

[0022] By generating control commands by matching a preset temperature control strategy library, an intelligent closed loop from temperature signal analysis to equipment regulation is achieved. This strategy library can cover optimized control schemes for forgings of different materials and sizes under various working conditions, enabling control commands to quickly adapt to specific processing needs, reducing reliance on manual experience, and improving the consistency and reliability of forging tempering treatment. Attached Figure Description

[0023] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent temperature control-based tempering treatment method for forgings described in this invention. Figure 2 A flowchart for acquiring multi-scale temperature signal components; Figure 3 Flowchart for determining the optimal temperature control frequency parameters; Figure 4 A flowchart for obtaining the steady-state and dynamic response components of an organization. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figure 1 This invention provides a method for tempering forgings based on intelligent temperature control, the method comprising: By monitoring the temperature changes of forgings in real time during heat treatment, and combining multi-scale signal decomposition, material phase transformation characteristic analysis, and intelligent optimization algorithms, the temperature control strategy is dynamically adjusted to achieve precise tempering. This method first acquires the real-time temperature time series and temperature distribution characteristics of the forgings. Based on the optimal signal decomposition basis function and decomposition depth level, the temperature time series is decomposed at multiple scales to extract multi-scale temperature signal components. The complexity metric of the decomposition coefficients is calculated to analyze their randomness characteristics, and key temperature region coefficients most relevant to the microstructure evolution of the forgings are selected. Combining the material phase transformation baseline characteristics, the current heat treatment stage state, and randomness characteristics, the optimal temperature control frequency parameters are determined. Based on the key temperature region coefficients and the optimal temperature control frequency parameters, the temperature field evolution characteristic sequence is reconstructed and divided into a gradual component reflecting the steady-state evolution of the microstructure and a rapid component reflecting the dynamic response of the phase transformation through modal decomposition. According to the tempering state type, a preset temperature control strategy library is matched to generate control commands for the heat treatment equipment, achieving dynamic and precise control.

[0026] Example 1: See Figure 2 In determining the optimal signal decomposition basis function and decomposition depth level, the support interval length and regularization order of candidate basis functions are selected as core function characteristic parameters. The support interval length determines the localization capability of the basis function in the time domain, while the regularization order affects the smoothness and noise resistance of the function. A swarm intelligence algorithm is used to process the function characteristic parameters, real-time temperature distribution characteristics, and a preset decomposition depth range. The swarm intelligence algorithm initializes multiple candidate basis function combinations, each containing a specific support interval length, regularization order, and decomposition depth level. The algorithm iteratively calculates the decomposition effect of each combination on the temperature time series, and the evaluation index includes the matching degree between the signal reconstruction error and the temperature distribution characteristics. The reconstruction error is quantified by the root mean square deviation between the decomposed signal and the original signal, and the feature matching degree is calculated by the similarity of key features such as temperature gradient distribution and extreme point positions. The algorithm selects the basis function with the smallest reconstruction error and the highest feature matching degree as the optimal signal decomposition basis function, and records its corresponding decomposition depth level. This level determines the number of scales for signal decomposition; too low a level may lead to loss of details, while too high a level introduces redundant noise.

[0027] A fast decomposition algorithm is employed, combining optimal signal decomposition basis functions with hierarchical decomposition depth to process real-time temperature time series. The fast decomposition algorithm is implemented using a filter bank, decomposing the signal step-by-step through low-pass and high-pass filter banks. In each decomposition level, the input signal is processed by a high-pass filter to extract detail components and a low-pass filter to extract approximate components, which are then used as input for the next decomposition level. During decomposition, a mirror extension method is used at sequence boundaries to suppress edge effects. Specifically, the original sequence is symmetrically copied and extended at both ends to form an extended sequence before further decomposition. After decomposition, the results for the corresponding intervals of the original sequence are truncated. This method effectively avoids spectral leakage and distortion caused by data discontinuities at boundaries. After decomposition, a set of decomposition coefficients at multiple scales is extracted, including high-frequency detail coefficients and low-frequency approximate coefficients. High-frequency detail coefficients reflect instantaneous temperature fluctuations and noise interference, while low-frequency approximate coefficients characterize the macroscopic trend of temperature changes. Based on the coefficient set, multi-scale temperature signal components are constructed, each corresponding to temperature characteristics at a specific time resolution: high-scale components (low frequency) exhibit slow evolution from hours to minutes, while low-scale components (high frequency) capture rapid fluctuations from seconds to milliseconds.

[0028] To extract the randomness characteristics of multi-scale temperature signal components, the complexity metric of each decomposition coefficient is calculated. The complexity metric is implemented using a multi-scale entropy algorithm. For the coefficient sequence at each scale, a new sequence with different scale factors is first generated through coarse-graining: the original sequence is divided into non-overlapping windows of length τ, and the mean within each window is calculated to form a coarse-grained sequence. Subsequently, the sample entropy value of each coarse-grained sequence is calculated; the sample entropy reflects the randomness and irregularity of the sequence. During calculation, an embedding dimension m and a tolerance threshold r are set, and the matching probability of vectors of length m and m+1 is statistically analyzed. The entropy value is obtained through negative logarithmic calculation. The multi-scale entropy curve (the change of entropy value with the scale factor τ) reveals the random distribution law of the temperature signal at different time scales. Abrupt entropy changes at high scales may correspond to critical points of material phase transitions, while entropy oscillations at low scales reflect local inhomogeneities in heat conduction.

[0029] Based on the multi-scale temperature signal components and their randomness, key temperature region coefficients most relevant to the microstructure evolution of forgings were selected. A strategy combining correlation analysis and feature importance assessment was adopted. First, the correlation coefficients between the time-domain features such as temperature gradients and extreme frequencies of each scale component and metallographic parameters, including offline detection data such as grain size and phase ratio, were calculated. Simultaneously, a random forest model was constructed, inputting features such as multi-scale entropy values ​​and decomposition coefficient statistics, and outputting microstructure prediction results. Key coefficients were then selected by ranking features by importance. Finally, coefficients with correlation coefficients greater than 0.7 and ranking in the top 20% of importance were selected as key temperature region coefficients. These coefficients are concentrated at specific scales (usually mid-to-high frequency scales), corresponding to the spatial coordinates and time nodes of active phase transformation regions in the temperature field.

[0030] Example 2: See Figure 3 This study obtained the thermal response characteristics of forging materials under different phase transformation states through multi-stage heat treatment experiments. The experimental design covered the entire heat treatment cycle, including three typical processes: heating, holding, and cooling. In the heating stage, the forging was heated to the target temperature at a constant or variable rate; in the holding stage, the target temperature was maintained and temperature fluctuation characteristics were recorded; in the cooling stage, different media such as air cooling, oil cooling, or water cooling were used for gradient cooling. During the experiment, thermocouple arrays were placed at key locations on the surface and inside the forging to collect temperature data in real time. Simultaneously, metallographic observation equipment was used to sample and analyze the microstructure at specific time points, establishing a three-dimensional mapping relationship between temperature, time, and microstructure evolution. Based on the experimental data, baseline characteristics of the material's phase transformation were extracted, including core parameters such as austenitization initiation temperature, pearlite transformation plateau range, and critical cooling rate for martensite transformation. These parameters constitute a benchmark database of the material's thermal response.

[0031] For real-time heat treatment processes, state parameters of the current heat treatment stage are extracted. By analyzing the first and second derivative characteristics of the temperature time series, the current stage (heating, holding, or cooling) is identified. The heating stage is characterized by a continuous temperature increase and a stable rate of change; the holding stage is characterized by temperature fluctuations remaining within a very small threshold; and the cooling stage is characterized by a monotonically decreasing temperature with the rate of change affected by the medium. The stage state is quantified as a multi-dimensional vector, including the current temperature value, stage duration, absolute value of the temperature change rate, and standard deviation of the fluctuation. Simultaneously, considering the randomness characteristics of multi-scale temperature signal components, non-stationarity analysis is performed on the temperature series. The rescaled range method is used to calculate the Hurst exponent to assess the long-range correlation of temperature fluctuations; wavelet variance analysis is used to identify the intensity distribution of random noise; and the permutation entropy algorithm is used to quantify the degree of irregularity of the sequence. These features collectively describe the random state of the temperature field, reflecting the uncertainties in the heat conduction process.

[0032] When determining the optimal temperature control frequency parameters, the frequency domain characteristics of the material's phase transformation baseline are first analyzed. Fourier transforms are performed on typical phase transformation temperature curves to extract frequency domain energy distribution characteristics. Dominant frequency components and their harmonic components are identified, and phase transformation-sensitive frequency bands are marked. For example, austenitization corresponds to low-frequency energy accumulation in the 0.05-0.2Hz range, while martensitic transformation corresponds to a high-frequency energy surge in the 1-5Hz range. Based on this, a candidate frequency set is established, covering a frequency range from 0.01Hz to 10Hz, with uniform sampling at logarithmic intervals. A three-layer fusion scoring calculation is performed on each candidate frequency: the phase transformation matching score is obtained by calculating the overlap area between the candidate frequency and the phase transformation-sensitive frequency band; the stage state fitness score is calculated based on the thermodynamic requirements of the current stage, for example, low frequencies are preferred in the heating stage to maintain thermal inertia, while high frequencies are preferred in the cooling stage to improve response speed; the randomness correlation score is determined by the Pearson correlation coefficient between the candidate frequency and the multi-scale entropy curve.

[0033] A genetic algorithm is used to dynamically optimize the scoring weights. A chromosome population containing weight coefficients is initialized, with each chromosome containing three gene segments: phase transition matching degree, stage fitness, and randomness association. The fitness function is set to maximize the variance of the candidate frequency's comprehensive score. The population is iteratively evolved through selection, crossover, and mutation operations. The calculation terminates when the optimal chromosome remains unchanged for five consecutive generations or the maximum number of iterations is reached, and the final weight coefficient combination is output. The weights are applied to the weighted sum of the three-layer scores to calculate the comprehensive score of the candidate frequencies. The frequency with the highest score is selected as the optimal temperature control frequency parameter, which simultaneously satisfies the triple requirements of phase transition characteristic matching, stage fitness, and randomness control. The parameter update mechanism is set to recalculate during stage switching or refresh periodically within a constant time period.

[0034] To verify the effectiveness of the parameters, a closed-loop feedback mechanism was constructed. The optimal temperature control frequency parameter was input into the PID controller, which adjusted the heating power or cooling medium flow rate in real time. The control effect of the frequency parameter was evaluated by comparing the dynamic error between the actual temperature curve and the theoretical phase transition curve. When the error continuously exceeded the tolerance threshold, the parameter recalculation process was triggered. This mechanism ensures that the temperature control frequency parameter always adapts to the dynamic characteristics of the material phase transition and real-time operating conditions.

[0035] Example 3: See Figure 4 In reconstructing the temperature field evolution feature sequence, a spatiotemporal mapping matrix is ​​first constructed as the core data structure for feature extraction. The row vectors of this matrix correspond to spatial sampling points, and the column vectors correspond to time nodes. The matrix elements are generated by a nonlinear combination of key temperature region coefficients, optimal temperature control frequency parameters, and phase transition modulation coefficients. The key temperature region coefficients are derived from the feature selection results after multi-scale decomposition, reflecting local features in the temperature field that are strongly correlated with microstructure evolution. The optimal temperature control frequency parameters characterize the most suitable control frequency band under the current heat treatment stage. The phase transition modulation coefficients are set based on the material's phase transition kinetics and are used to balance the energy distribution at different phase transition stages. The generation of the matrix elements follows the following nonlinear relationship:

[0036] In the formula: This represents the value of the element in the i-th row and j-th column of the spatiotemporal mapping matrix. Let (i,j) be the critical temperature region coefficient corresponding to the location (i,j). It is the optimal temperature control frequency parameter at time node j. This is the reference frequency offset. It is the modulation coefficient in the current phase transition stage. This is the amplitude scaling factor. Controlling the steepness of the frequency response, The phase transition modulation intensity is adjusted. This formula achieves frequency selectivity enhancement through the sigmoid function, constrains the phase transition modulation range through the hyperbolic tangent function, and finally outputs a feature matrix with spatiotemporal correlation.

[0037] Multi-scale time-series analysis was performed on the spatiotemporal mapping matrix using a three-level analysis architecture: The short-time analysis window, 5-10 seconds long, extracts local frequency domain features using sliding window Fourier transform, focusing on capturing instantaneous frequency shifts and energy redistribution during phase transitions; the medium-time analysis window, covering 1-5 minutes, calculates the mean, variance, and autocorrelation function of matrix elements within the window to track the evolution trend of the thermal field's steady state; the long-time analysis window spans the entire heat treatment stage, employing singular spectral analysis to decompose the spatiotemporal modes of the matrix and monitor the energy accumulation of low-frequency dominant components during tissue evolution. The results of the three levels of analysis generate short-time transition feature vectors, medium-time steady-state feature matrices, and long-time trend tensors, respectively.

[0038] A temperature field evolution feature sequence is generated by adaptively weighting and fusing multi-scale analysis results. Weights are calculated based on confidence indices for each scale: the confidence for short-time analysis is determined by the frequency domain signal-to-noise ratio, the confidence for medium-time analysis depends on the decay rate of the autocorrelation function, and the confidence for long-time analysis is assessed by the concentration of modal energy. The fusion process employs three-dimensional tensor stitching and dimensionality reduction to project features from different scales onto a unified time axis, forming a feature sequence containing multi-modal phase transition characteristics. Each time point in this sequence contains three types of information: spatial distribution characteristics, frequency domain characteristics, and evolution trend, achieving a complete characterization of the dynamic properties of the temperature field.

[0039] When performing mode decomposition on the temperature field evolution feature sequence, an improved extremum detection algorithm is used to locate feature extrema. The algorithm introduces an adaptive threshold mechanism, dynamically adjusting the extremum determination threshold based on the local fluctuation amplitude of the sequence: reducing threshold sensitivity in stable segments to avoid missing weak features, and increasing threshold noise resistance in segments with sharp fluctuations to prevent spurious extremum interference. After detecting an extremum, the sequence boundary is processed through a mirror-linear hybrid extension: the left boundary extends the data before the first extremum using mirror symmetry, and the right boundary extends linearly after the last extremum. The extended sequence is then subjected to an iterative screening process to extract intrinsic mode components. In each iteration, the mean curves of the upper and lower envelopes are calculated, and the new sequence is obtained by subtracting the mean curve from the original sequence, until the envelope symmetry and zero-crossing equilibrium conditions are satisfied.

[0040] The multi-mode components obtained from the decomposition are classified according to their average period and energy distribution. When calculating the average period of each component, a weighted average with zero-crossing intervals is used, with the weight determined by the amplitude difference between adjacent extreme points. The energy distribution is evaluated using the integral area of ​​the component power spectrum in the 0.1-10Hz frequency band. The classification rules are as follows: components with an average period less than 30 seconds and energy concentrated above 1Hz are classified as rapid-change components, reflecting the rapid phase transformation response during quenching; components with an average period exceeding 2 minutes and energy concentrated below 0.5Hz are classified as slow-change components, characterizing the steady-state adjustment of the microstructure during annealing; intermediate-state components are assigned based on their coherence with the rapid / slow-change components. The residual terms, containing trend information and nonlinear drift, are included in the slow-change component category.

[0041] The separation of rapid and slow-changing components is achieved through independent channel analysis. Hilbert transform is applied to the rapid component to obtain its instantaneous frequency and amplitude, constructing a phase transition dynamic response spectrum. Frequency abrupt changes in the spectrum correspond to key events such as the initiation of martensitic transformation. Moving average smoothing is applied to the slow-changing component, and its time-domain envelope curve is extracted. Inflection points on the curve mark steady-state transition nodes such as the completion of recrystallization. The time-frequency characteristics of both types of components together constitute the dual-channel input for tempering state identification, providing differentiated feature basis for subsequent control strategy generation.

[0042] Example 4: In identifying the quenching and tempering state type of forgings, the gradual change components are first subjected to feature extraction. These components reflect the steady-state evolution trend of the microstructure, and their feature extraction mainly focuses on time-domain statistical characteristics, frequency band energy distribution, and inter-component correlations. Time-domain statistical characteristics include mean, variance, skewness, and kurtosis; these parameters describe the central tendency, dispersion, and distribution pattern of temperature changes. Frequency band energy distribution characteristics are obtained by decomposing the signal into different frequency bands and calculating the energy proportion, with a focus on the energy accumulation in the 0.01-0.1Hz low-frequency band. Inter-component correlation characteristics calculate the cross-correlation coefficients between each gradual change component, reflecting the synchronicity of temperature evolution at different spatial locations.

[0043] Taking a specific forging process as an example, Table 1 shows the feature extraction results of the gradient component within a certain time period. The forging material is 42CrMo steel, and the heat treatment stage is the late holding stage. The feature data comes from the monitoring results of three spatial sampling points (P1, P2, P3) and five time windows (T1-T5).

[0044] Table 1: Example data for feature extraction of gradual change components.

[0045]

[0046] Based on the data in Table 1, steady-state feature vectors were constructed. Each time window's feature vector contains 15 dimensions (3 sampling points × 5 feature parameters). A hierarchical classification strategy was employed to process these feature vectors. The first level used K-means clustering to divide the steady-state features into three main categories: uniform steady-state, gradient steady-state, and fluctuating steady-state. Uniform steady-state is characterized by similar mean temperatures at each sampling point, small variance, and high cross-correlation coefficients; gradient steady-state shows spatial temperature differences but remains stable over time; and fluctuating steady-state exhibits periodic small fluctuations. The second level of classification used support vector machines to refine the state subclasses, for example, further dividing gradient steady-state into axial gradient and radial gradient, etc.

[0047] Feature extraction of rapid change components focuses on transient characteristic analysis. Transient amplitude rate is calculated to represent the instantaneous slope of temperature change, identifying abrupt changes in rapid rise or fall. Peak characteristic analysis includes three parameters: peak amplitude, rise time, and fall time, used to quantify the energy release characteristics during phase transitions. Waveform structure features describe the morphological regularity of transient events through zero-crossing intervals and waveform factors. These features constitute a transition feature vector, with each transient event corresponding to a 12-dimensional feature combination.

[0048] The threshold detector employs a dynamic threshold mechanism when processing transition eigenvectors. For the peak amplitude parameter, the threshold is dynamically adjusted with each heat treatment stage: a higher threshold is set during the heating stage to filter noise interference, while the threshold is lowered in the critical phase transformation region to improve sensitivity. The threshold ratio of rise time to fall time is preset based on the material's phase transformation kinetics; for example, the rise time threshold for austenitization is set to 3-5 seconds, and the fall time threshold for martensitic transformation is set to 1-3 seconds. When an eigenvector exceeds any threshold, it is marked as a valid phase transformation event, and its type and intensity are recorded.

[0049] When fusing steady-state classification results with phase transition events using a probabilistic network, a two-layer network structure is constructed. The upper layer processes the combined probabilities of steady-state types and phase transition events, while the lower layer evaluates the reliability of spatiotemporal continuity features. Spatiotemporal continuity checks include the reasonableness of time intervals in the event sequence, consistency of spatial propagation, and energy conservation verification. For example, martensitic transformation events should follow a spatial pattern of propagation from the surface to the core, and the time intervals between adjacent events should match the cooling rate. The network output includes five basic tempering state types: homogeneous austenitization, localized overheating, normal quenching, incomplete quenching, and over-tempering, each accompanied by a confidence score of 0-1.

[0050] Taking the processing of the data in Table 1 as an example, the system detected the following key characteristics: the gradual change component exhibits uniform steady-state characteristics (cross-correlation coefficient > 0.9, variance < 6), and the rapid change component captures a transient event with an amplitude of 15℃ and a rise time of 4 seconds within the T2 time window. Combined with material database information, this event characteristic matches the austenitization completion marker of 42CrMo steel. After probabilistic network calculation, the system outputs a "uniform austenitization" state type with a confidence level of 0.87. Based on this determination, the system automatically triggers the next stage of quenching and cooling command, initiating the oil cooling cycle and setting the initial cooling rate to 25℃ / s.

[0051] The entire state recognition process operates using a sliding window mechanism, updating the feature extraction and classification results every 10 seconds. When the output states of three consecutive windows are consistent and the confidence level is above 0.8, a stable state is determined, and control parameters are locked. If a state transition occurs or the confidence level falls below 0.6, a review process is initiated to reanalyze historical data. This design ensures timely response to rapid phase transitions while avoiding misjudgments caused by transient interference.

[0052] Example 5: During the matching process with the preset temperature control strategy library, the system maps the identified tempering state types to pre-stored control parameter templates. The strategy library adopts a tree-like index structure. The first level is classified according to material grade, the second level is divided according to the geometric complexity of the forging, and the third level corresponds to the specific tempering state type. Each state type is associated with a set of temperature curve templates, holding time parameters, and cooling rate constraints. The temperature curve templates include three basic modes: the linear heating template sets the initial temperature, target temperature, and constant heating rate; the stepped holding template defines multiple holding platforms and their holding times; and the gradient cooling template specifies the initial cooling rate, inflection point, and rate change function. The holding time parameter is set based on experimental data of material phase transformation kinetics. For carbon steel materials, it is usually set to 0.8-1.2 minutes per millimeter of thickness, while for alloy steel, it is extended to 1.5-2 minutes. The cooling rate constraints include the maximum allowable rate and the minimum critical rate to prevent cracking due to excessively rapid cooling or coarsening of the microstructure due to excessively slow cooling.

[0053] The control command generation employs a parameter synthesis mechanism. The system first selects a temperature profile template matching the current state; for example, when an "incomplete quenching" state is identified, a secondary quenching template is automatically selected. This template contains two characteristic segments: the first stage cools at a high rate to 50°C below the martensitic transformation initiation temperature; the second stage reduces the cooling rate and holds the temperature to complete the residual austenite transformation. The holding time parameter is dynamically calculated based on the actual thickness of the forging, using phase transformation latency data from the material database as a benchmark, multiplied by a thickness correction factor. The cooling rate constraint is checked in real-time using a thermal stress model; when the calculated stress approaches the material's yield strength, the upper limit of the rate is automatically reduced. The final generated control command contains three sets of core parameters: the target temperature value is accurate to ±5°C; the heating slope is set in stages in °C / min; and the stage switching condition uses a composite triggering mechanism, including both temperature threshold triggering and setting a maximum time protection limit.

[0054] After the command is executed, a real-time monitoring process is initiated. An infrared thermal imager array is deployed on the surface of the forging to acquire temperature distribution cloud maps at a sampling frequency of 10Hz; thermocouple groups are implanted at key internal locations to monitor the temperature change trend of the core. Three characteristic quantities are calculated based on the surface temperature distribution analysis: radial temperature difference extreme value, circumferential temperature non-uniformity, and surface-to-core temperature gradient. Internal heat flow characteristics are obtained by solving the inverse problem of heat conduction, and a three-dimensional transient model is established to calculate the heat flux density distribution. The deviation metric is calculated using a dual-channel approach: the temperature field deviation channel calculates the spatial root mean square error between the current temperature distribution and the target curve; the heat flow deviation channel assesses the directional consistency between the actual heat flow and the theoretical value. The two are weighted and fused to form a comprehensive deviation index, with the weighting coefficient dynamically adjusted according to the heat treatment stage. The heating stage emphasizes temperature field deviation, while the cooling stage prioritizes heat flow deviation.

[0055] When the overall deviation index exceeds a preset threshold, the control command iterative update process is triggered. The threshold setting adopts an adaptive mechanism: a relatively lenient threshold is set in the initial stage (e.g., 15% relative error), and it is gradually tightened to 5% as the processing progresses. The update process first re-acquires real-time temperature time series, increasing the sampling frequency to 1.5 times the original value to capture transient details. The temperature distribution feature update includes three improvements: spatial grid refinement, locally refining high-gradient regions; time series segmentation and reorganization, dividing key analysis intervals according to deviation distribution characteristics; and feature extraction parameter optimization, enhancing sensitivity to abrupt change regions. Based on the updated features, the multi-scale decomposition, key coefficient screening, and temperature control parameter calculation processes are re-executed.

[0056] The iterative adjustment employs a gradual convergence strategy. The initial adjustment modifies only the cooling rate parameter, keeping other parameters unchanged. If the second monitoring still shows deviations, the heating slope and holding time are adjusted simultaneously. Three consecutive deviations trigger a full parameter regeneration. Each adjustment records the parameter modification amount and deviation improvement rate, forming a closed-loop learning sample. After more than five processing cycles of forgings of the same material, the system automatically updates the basic parameter templates in the strategy library, gradually improving the initial control accuracy. The entire temperature control process, through this perception-decision-execution-feedback loop mechanism, achieves dynamic optimization of heat treatment parameters, adapting to the tempering requirements under different operating conditions.

[0057] The transmission and execution of control commands are protected by multiple safeguards. Commands are issued using a three-way handshake protocol; the device controller immediately sends back a parameter verification code upon receipt. The execution process involves step-by-step confirmation: the heating system sends a ready signal after completing the target temperature setting, and the cooling system performs a media flow self-check before startup. The execution status is monitored in real time; if the actual heating rate deviates from the command value by more than 10% for 30 seconds, or if the cooling rate fluctuation exceeds the permissible range, a pause protection mechanism is automatically triggered, and an error code is reported. This design ensures reliable execution of control commands and provides fault-tolerant protection for the system.

[0058] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0059] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for tempering forgings based on intelligent temperature control, characterized in that, include: The real-time temperature time series and corresponding temperature distribution characteristics of the forging during the heat treatment process are obtained, and the optimal signal decomposition basis function and decomposition depth level corresponding to the temperature distribution characteristics are obtained. These are used to perform multi-scale signal decomposition on the real-time temperature time series to obtain multi-scale temperature signal components. Obtain the complexity metric of the decomposition coefficients corresponding to the multi-scale temperature signal components in order to extract the randomness characteristics of the decomposition coefficients; Based on the multi-scale temperature signal components and randomness characteristics, the key temperature region coefficients most relevant to the evolution of forging microstructure are selected from multiple decomposition coefficients. The baseline characteristics of material phase transformation, the current heat treatment stage state, and the randomness characteristics of the forging are obtained to determine the optimal temperature control frequency parameters that are suitable for the material properties. Based on the key temperature region coefficient and the optimal temperature control frequency parameter, a temperature field evolution characteristic sequence is reconstructed; mode decomposition is performed on the temperature field evolution characteristic sequence to obtain the slow-change component reflecting the steady-state evolution of the tissue and the rapid-change component reflecting the dynamic response of the phase transition. Based on the slow-change components of the steady-state evolution of the structure and the rapid-change components of the phase transformation dynamic response, the corresponding tempering state type of the forging is identified. According to the tempering state type, a preset temperature control strategy library is matched to generate control instructions for the heat treatment equipment to achieve tempering of the forging.

2. The method according to claim 1, characterized in that, The step of obtaining the optimal signal decomposition basis function and decomposition depth level corresponding to the temperature distribution characteristics includes: Select the functional characteristic parameters of the candidate basis functions, wherein the functional characteristic parameters include at least the support interval length and the regularization order; By inputting the function characteristic parameters, the temperature distribution features, and the preset decomposition depth range through a swarm intelligence algorithm, the optimal signal decomposition basis function and its corresponding decomposition depth level are determined from the candidate basis functions.

3. The method according to claim 1, characterized in that, The step of performing multi-scale signal decomposition on the real-time temperature time series to obtain multi-scale temperature signal components includes: The real-time temperature time series is processed using a fast decomposition algorithm based on the optimal signal decomposition basis function and decomposition depth level; edge effects are suppressed at the sequence boundaries using a mirror continuation method. Extract a set of decomposition coefficients at multiple scales, and construct the multi-scale temperature signal components based on the set of decomposition coefficients.

4. The method according to claim 1, characterized in that, The acquisition of the material phase transformation baseline characteristics, current heat treatment stage state, and randomness characteristics of the forging includes: The thermal response characteristics of materials under different phase transition states were obtained through multi-stage heat treatment experiments, wherein the multi-stage process includes at least a heating stage, a holding stage, and a cooling stage. Based on the material's thermal response characteristics, the current heat treatment stage state and randomness characteristics are extracted.

5. The method according to claim 4, characterized in that, The determination of the optimal temperature control frequency parameters that adapt to the material properties includes: Based on the frequency domain energy distribution characteristics, current heat treatment stage state, and randomness characteristics of the material phase transition baseline features, a multi-layer fusion scoring calculation is performed on the preset candidate frequency set. The scoring includes phase transition matching score, stage state fitness score, and randomness correlation score. The weighting coefficients of the score are dynamically adjusted using an optimization algorithm to determine the optimal temperature control frequency parameter from the candidate frequency set.

6. The method according to claim 1, characterized in that, The step of reconstructing and generating a temperature field evolution characteristic sequence based on the key temperature region coefficient and the optimal temperature control frequency parameter includes: A spatiotemporal mapping matrix is ​​constructed, wherein the matrix elements of the spatiotemporal mapping matrix are generated by nonlinear combination of the key temperature region coefficient, the frequency matching degree corresponding to the optimal temperature control frequency parameter, and the preset phase change modulation coefficient. Multi-scale time series analysis is performed on the spatiotemporal mapping matrix. The multi-scale time series analysis includes short-time analysis to capture phase transition characteristics, medium-time analysis to track the thermal field stability state, and long-time analysis to monitor tissue evolution trends. By adaptively weighting and fusing the results of the multi-scale time series analysis, a temperature field evolution feature sequence containing multimodal phase transition characteristics is generated.

7. The method according to claim 1, characterized in that, The step of performing mode decomposition on the temperature field evolution characteristic sequence to obtain the slow-change component reflecting the steady-state evolution of the tissue and the rapid-change component reflecting the dynamic response of the phase transition includes: The components of the temperature field evolution feature sequence are subjected to extreme point detection and boundary extension processing, and multi-mode components that satisfy the intrinsic conditions are obtained through iterative screening. Based on the average period and energy distribution characteristics of the multi-mode components, the earlier short-period components are classified as rapid-change components, and the later long-period components and residual terms are classified as slow-change components.

8. The method according to claim 1, characterized in that, The identification of the quenching and tempering state type corresponding to the forging includes: Extract the time-domain statistical features, frequency band energy distribution features, and inter-component correlation features of the gradual variation components to construct a steady-state feature vector; The transient amplitude change rate, peak characteristics and waveform structure features of the rapid change component are extracted to construct a transition feature vector; A hierarchical classification strategy is used to identify the state of the steady-state feature vector, and a threshold detector is used to identify the phase transition event corresponding to the transition feature vector. Based on the fusion of the state recognition results and phase transition events using a probabilistic network, and combined with spatiotemporal continuity features, confidence assessment is performed to output the heat treatment state type.

9. The method according to claim 1, characterized in that, The preset temperature control strategy library is matched to generate control commands for the heat treatment equipment, including: The heat treatment state type is mapped and matched with the temperature curve template, heat preservation time parameter and cooling rate constraint in the strategy library; The control command, which includes the target temperature value, the heating slope, and the stage switching conditions, is generated based on the matching results.

10. The method according to claim 9, characterized in that, The control commands for generating the heat treatment equipment also include: Real-time monitoring of surface temperature distribution and internal heat flow characteristics of forgings, and calculation of the deviation metric between the current temperature field and the target temperature curve; When the deviation metric exceeds a preset threshold, the real-time temperature time series is reacquired and the temperature distribution characteristics are updated to iteratively adjust the control commands.

Images