An optimization method for an automobile casting forming process

Abstract

This invention relates to the field of casting process optimization technology, and more specifically, to an optimization method for automotive casting forming process. The method includes: acquiring a multi-channel standardized temperature discrete sequence during the casting solidification process; extracting local rank sequences using a one-dimensional sliding window at multiple scales, calculating the normalized gradient amplitude at each sampling point, and fusing them using a geometric confidence modulation factor generated by the inter-scale dispersion; obtaining the normalized oscillation based on the temperature change direction reversal characteristics within the symmetrical local window, and attenuating and correcting the fused gradient to obtain the final gradient amplitude; determining the actual crystallization plateau duration for each channel based on the final gradient amplitude to adjust the mold cooling water flow rate. This invention improves the stability of phase transition boundary identification and reduces incorrect cooling water adjustment.

Description

Technical Field

[0001] This invention relates to the field of casting process optimization technology. More specifically, this invention relates to an optimization method for the forming process of automotive castings. Background Technology

[0002] With the increasing demands for lightweighting in the modern automotive industry, large and complex thin-walled aluminum alloy castings are being used more and more widely in automobile manufacturing. In the process of forming automotive castings, the solidification and cooling stage after the molten metal is injected into the mold is a crucial step that determines the final microstructure and mechanical properties of the casting. Commonly used casting materials such as aluminum alloys have a wide crystallization temperature range during solidification. When they cool and pass through the liquidus and solidus lines, the crystallization and precipitation of the microstructure are accompanied by the release of a large amount of latent heat of phase transformation. This latent heat of phase transformation is manifested on the macroscopic cooling curve as a significant slowdown in the rate of temperature decrease, forming a characteristic region of abrupt slope change or a plateau period. Accurately locating the start and end points of this plateau period is crucial for real-time assessment of whether defects such as shrinkage porosity and cold shuts occur inside the casting, and for dynamically adjusting forming process parameters such as the mold cooling water flow rate accordingly.

[0003] Currently, one-dimensional mathematical morphological feature extraction methods are commonly used in industrial data processing to automatically extract phase transition boundaries from temperature time-series signals. This method defines a fixed-length one-dimensional linear structuring element, performs sliding operations on the cooling curve, uses an expansion operator to extract local temperature maxima, and an corrosion operator to extract local temperature minima, subtracting the two to obtain the morphological gradient. Since the local maxima and minima are very close during the crystallization plateau, their morphological gradient rapidly approaches zero; existing techniques use this to locate gradient troughs and pinpoint phase transition boundaries.

[0004] However, applying this method to automotive die casting or sand casting workshops has serious limitations. The automotive casting environment is harsh, with mechanical impacts from large mold clamping mechanisms, high-frequency vibrations from injection punches, electromagnetic radiation within the mold, and poor contact in thermocouple circuits all superimposing high-frequency spike noise onto the real-time temperature curve. Since the basic morphological expansion and corrosion operators are mathematically essentially local absolute extremum searches, this mechanism has a natural amplification effect on high-frequency spike noise. When structural elements slide over isolated temperature burrs caused by electromagnetic interference, the operator instantly captures abnormal extrema, causing huge pseudo-gradient spikes to erupt on the smooth morphological gradient curve. This masks the weak real gradient changes during the release of latent heat of crystallization, leading to frequent misjudgments of phase transition boundaries in the molding process control system, triggering false defect alarms, and causing the mold cooling system to start and stop at incorrect times. This, in turn, leads to thermal stress concentration or internal thermal cracks in the casting, severely restricting the improvement of the yield rate of high-quality automotive castings. Summary of the Invention

[0005] To address the problem that existing morphological gradient methods are susceptible to high-frequency spike noise interference, leading to misjudgment of phase transformation boundaries and cooling control failure in castings, this invention proposes an optimization method for automotive casting forming processes. This method includes the following steps: Multi-channel temperature signals during the solidification process of castings are collected, and after synchronization alignment, baseline compensation and geometric normalization mapping, a standardized temperature discrete sequence for each channel is obtained. For any channel's standardized temperature discrete sequence, a linear sliding window of multiple scales is used to slide through the sample points in unit step sizes. For each sample point, the local rank sequence of each sample point at each scale is obtained based on the temperature magnitude relationship of the sample points within the window. Based on the local rank sequence, the normalized gradient magnitude of each sample point at each scale is obtained. Based on the normalized gradient magnitude of the same sampling point at each scale, the geometric confidence modulation factor of each sampling point is obtained; the product of the geometric confidence modulation factor and the sum of the normalized gradient magnitudes of each sampling point at all scales is used as the multi-scale fused morphological gradient magnitude of each sampling point; based on the flipping feature of the temperature change direction within the symmetrical local window of each sampling point, the normalized oscillation of each sampling point is obtained; the normalized oscillation is used to correct the multi-scale fused morphological gradient magnitude to obtain the final gradient magnitude of each sampling point. The actual duration of the crystallization platform in each channel is determined based on the final gradient amplitude. The duration of the channel with the slowest solidification is compared with the standard crystallization process reference duration, and the flow rate of the mold cooling water is adjusted accordingly.

[0006] This invention standardizes multi-channel temperature signals and combines them with multi-scale ergonomics to obtain normalized gradient amplitude, geometric confidence modulation, and oscillation correction. This effectively suppresses interference from high-frequency spike noise, scale mismatch, and temperature fluctuations on phase transition boundary identification, improves the accuracy of judging the duration of the crystallization platform in automotive castings, thereby reducing misadjustment of mold cooling water flow and improving casting quality and process stability.

[0007] Preferably, obtaining the local rank sequence of each sampling point at each scale includes: The preset short-scale window length is N, the medium-scale window length is 2N, and the long-scale window length is 3N. For any channel's standardized temperature discrete sequence, a one-dimensional linear sliding window is established with the short-scale window length N, the medium-scale window length 2N, and the long-scale window length 3N. The one-dimensional linear sliding window at each scale is slid forward along the standardized temperature discrete sequence in unit steps. At each sampling point, all temperature sampling points within the window are rearranged in ascending order of value to obtain the local rank sequence of each sampling point at each scale.

[0008] By setting different window sizes and performing sliding sorting on the temperature sequences, it is possible to simultaneously adapt to the different thermal response characteristics of rapid solidification in thin-walled regions and slow solidification in thick-walled regions, avoiding excessive smoothing or spurious oscillations at phase boundaries caused by a single window size, and improving the adaptability of temperature feature extraction for different structural regions.

[0009] Preferably, obtaining the normalized gradient magnitude of each sampling point at each scale includes: Obtain the local maximum, local minimum, and median rank reference values ​​for each sampling point at a short scale; In the formula, This represents the normalized gradient magnitude at the m-th sampling point under a short scale. This represents the local maximum value of the m-th sampling point at a short scale. This represents the local minimum value at the m-th sampling point under a short scale. This represents the median rank reference value of the m-th sampling point at a short scale. This represents a very small constant.

[0010] Obtain the normalized gradient magnitude of each sampling point at each scale to facilitate subsequent location of the phase transition boundary.

[0011] Preferably, obtaining the local maximum, local minimum, and median rank reference value of each sampling point at a short scale includes: Preset the high quantile rank coordinate parameter H and the low quantile rank coordinate parameter L; calculate the product of the short-scale window length N with H, L and 0.5, and round up to obtain the high quantile index value, low quantile index value and median index value; take the value at the position corresponding to the high quantile index value in the local rank sequence of each sampling point under the short scale as the local maximum value of each sampling point under the short scale; take the value at the position corresponding to the low quantile index value in the local rank sequence of each sampling point under the short scale as the local minimum value of each sampling point under the short scale; take the value at the position corresponding to the median index value in the local rank sequence of each sampling point under the short scale as the median rank reference value of each sampling point under the short scale.

[0012] Instead of directly using the absolute maximum and absolute minimum values ​​within the window, the local maximum and local minimum values ​​are determined by using the high quantile rank coordinate parameter H and the low quantile rank coordinate parameter L. This can exclude isolated spikes, burrs and other abnormal sampling points from gradient calculation, and significantly reduce the amplification effect of electromagnetic interference, mechanical vibration and sensor transient anomalies on morphological gradients.

[0013] Preferably, obtaining the geometric confidence modulation factor for each sampling point includes: The arithmetic mean of the normalized gradient magnitudes of each sampling point at all scales is taken as the normalized gradient mean of each sampling point; the square of the deviation of the normalized gradient magnitude of each sampling point at each scale from the normalized gradient mean is calculated, and the square root of the arithmetic mean of the deviation squares is taken to obtain the dispersion index of each sampling point. In the formula, The geometric confidence modulation factor represents the current sampling point; This represents the dispersion index of the current sampling point; This represents the normalized gradient mean of the current sampling point; This represents a very small constant.

[0014] Preferably, obtaining the normalized oscillation value of each sampling point includes: For each sampling point on the standardized temperature discrete sequence of any channel, take M sampling points in front of and behind that point to form a symmetrical local window; obtain the orientation state sequence of the symmetrical local window for each sampling point; For the direction state sequence of the symmetrical local window at each sampling point, the total number of times the direction state flag changes from +1 to -1 and from -1 to +1 in the sequence is counted, which is used as the flip count of the symmetrical local window at each sampling point; the flip count of the symmetrical local window at each sampling point is divided by the maximum value of the flip counts of the symmetrical local windows at all sampling points to obtain the normalized oscillation of the symmetrical local window at each sampling point.

[0015] By constructing symmetrical local windows around each sampling point and counting the number of reversals in the direction of temperature change to obtain the normalized oscillation quantity, it is possible to effectively identify repeated up-and-down fluctuations in the temperature curve caused by cooling medium pulsation, thermal convection disturbance, or sampling jitter, providing a reliable basis for distinguishing between the true phase transition boundary and oscillation interference. Preferably, obtaining the directional state sequence of the symmetrical local window for each sampling point includes: The sampling points in the symmetrical local window of each sampling point are traversed one by one. The temperature value of the sampling point is subtracted from the temperature value of the subsequent adjacent sampling points of any sampling point to obtain the temperature difference value of the sampling point. If the temperature difference value is greater than zero, the sampling point is determined to be in a temperature rising state and is assigned a direction state mark of +1. If the temperature difference value is less than zero, the sampling point is determined to be in a temperature falling state and is assigned a direction state mark of -1. If the temperature difference value is equal to zero, the sampling point is determined to be in a temperature level state and is assigned a direction state mark of 0. The direction state marks of all sampling points in the symmetrical local window of each sampling point are arranged in chronological order of traversal to form a direction state sequence.

[0016] Preferably, the step of correcting the multi-scale fused morphological gradient magnitude using the normalized oscillation to obtain the final gradient magnitude of each sampling point includes: Subtract the normalized oscillation value of each sampling point from 1 to obtain the retention coefficient of each sampling point. Multiply the multiscale fused morphological gradient magnitude of each sampling point by the retention coefficient of each sampling point to obtain the final gradient magnitude of each sampling point.

[0017] Preferably, determining the actual duration of the crystallization platform in each channel includes: The maximum value of the final gradient amplitude of all sampling points in the current standardized temperature discrete sequence is taken as the global peak amplitude. Eight percent of the global peak amplitude is taken as the entry threshold, and ten percent of the global peak amplitude is taken as the exit threshold. A preset number of continuous sampling points K is used to traverse each sampling point in the current standardized temperature discrete sequence. If the final gradient amplitude of the sampling point and its subsequent K sampling points are all less than the entry threshold, the time corresponding to the sampling point is the phase boundary start time. Continue to traverse along the sampling point. If the final gradient amplitude of any sampling point and its subsequent K sampling points is greater than the exit threshold, the time corresponding to the sampling point is taken as the phase boundary end time. The difference between the phase boundary end timestamp and the phase boundary start timestamp is taken as the actual crystallization platform duration of the current channel.

[0018] Preferably, the step of comparing the time corresponding to the slowest solidification channel with the standard crystallization process reference time and adjusting the mold cooling water flow rate includes: The maximum value among the actual crystallization plateau durations of each channel is taken as the final crystallization plateau duration of this casting. The cooling deviation value is obtained by subtracting the final crystallization plateau duration of this casting from the preset standard crystallization process reference duration. When the cooling deviation value is positive, the cooling water flow rate is increased to accelerate heat dissipation. When the cooling deviation value is negative, the cooling water flow rate is decreased to mitigate thermal shock.

[0019] The present invention has the following beneficial effects: It transforms the multi-channel temperature curves of the solidification process of automotive castings into a standardized sequence under a unified scale. By using high-quantile and low-quantile rank coordinate parameters to determine local maxima and minima, it can exclude isolated peaks, burrs, and other abnormal sampling points from gradient calculation, significantly reducing the amplification effect of electromagnetic interference, mechanical vibration, and sensor transient anomalies on morphological gradients. Combined with geometric confidence modulation factor adjustment and direction reversal oscillation attenuation, the final gradient amplitude of each sampling point is obtained, avoiding the interference of scale mismatch and continuous jitter on phase transition boundary identification. Finally, based on the final gradient amplitude, the duration of the crystallization platform of each channel is obtained, and the cooling water flow is driven by the slowest solidification region in a closed loop, improving the solidification control accuracy of castings and reducing the risks of shrinkage porosity, thermal cracking, and false alarms. Attached Figure Description

[0020] Figure 1 This is a flowchart of the steps of an optimized method for automobile casting forming process according to an embodiment of the present invention; Figure 2 This is a schematic diagram of symmetrical local window construction. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0022] Please see Figure 1 The diagram illustrates a flowchart of an optimized method for automotive casting forming process according to an embodiment of the present invention, the method comprising the following steps: S001. Acquire multi-channel temperature signals during the solidification process of the casting, and obtain standardized temperature discrete sequences for each channel through synchronous alignment, baseline compensation, and geometric normalization mapping.

[0023] It should be noted that automotive castings do not solidify simultaneously at every location; the cooling conditions vary greatly in different areas. Therefore, it is necessary to collect temperature sequences from different areas to obtain the cooling status of different key areas of the casting.

[0024] In this embodiment of the invention, firstly, a multi-channel thermocouple sensor array arranged at key sampling points in the cavity of the automotive casting mold is used to collect the original simulated temperature electrical signals during the solidification process of the casting; and then a high-speed analog-to-digital conversion interface is used to synchronously sample and discretize the electrical signals of each channel. Using the hardware trigger pulse of the injection action as the zero-time anchor point, the data of each channel are strictly aligned to the same clock reference sequence axis in the time dimension, eliminating time axis misalignment caused by sampling clock drift between channels or injection response delay. Secondly, the temperature discrete sequences of each channel are subjected to one-dimensional median filtering to remove out-of-limit random code points caused by transient circuit breaks in the sensor circuit. Next, the steady-state temperature data sequence of each channel before the injection start is extracted. The mean of all data in the steady-state temperature data sequence of each channel is used as the baseline temperature reference value of each channel for the current casting. The baseline temperature reference value of the corresponding channel for the current casting is subtracted from each data in the temperature discrete sequence of each channel to obtain the true temperature discrete sequence of each channel. This is to compensate for the longitudinal baseline offset caused by the difference in smelting batches, the drift of ambient temperature and the residual error of thermocouple cold junction compensation, so that the cooling curves collected in different furnaces and different seasons have a unified longitudinal starting reference. Finally, a dimensionless geometric normalization mapping is performed on the true temperature discrete sequences of each channel. Specifically, the standard deviation of the steady-state temperature data sequence of each channel is obtained as the standard deviation of each channel. The true temperature discrete sequence of each channel is then divided point by point by the standard deviation of the corresponding channel to obtain the normalized temperature discrete sequence of each channel. This ensures that subsequent morphological operators still have a consistent geometric scale under cross-channel conditions, eliminating the amplitude magnitude dispersion caused by differences in sensor sensitivity.

[0025] S002. Preset multiple scale windows to slide through the current standardized temperature discrete sequence and obtain the morphological gradient sequence of each scale.

[0026] It should be noted that by defining a local window of fixed length and sliding the calculation on the cooling curve, the maximum and minimum temperature values ​​are extracted in each local window, and the morphological gradient is obtained by subtracting the two. Since the maximum and minimum temperature values ​​are very close during the crystallization plateau, the morphological gradient approaches zero sharply. The phase transition boundary is locked by finding the gradient trough region. However, isolated high-frequency spike noise induced by large mechanical vibration and electromagnetic radiation is common in die-casting workshops. When the structural element slides over these spikes, the absolute extremum search mechanism amplifies their values ​​by a factor of two, causing a drastic pseudo-abrupt change in the output gradient curve, which can easily mask the true weak phase transition geometric features. Moreover, conventional filtering cannot completely eliminate the noise mixed with the low-frequency band of the phase transition. Therefore, this invention designs a method where, when the local window slides to a certain position, the temperature values ​​in the local window are sorted in ascending order, and the temperature values ​​in the middle of the sequence are used to replace the maximum and minimum temperature values ​​in the local window. This isolates the noise spikes at both ends of the sequence so that they do not participate in the gradient calculation, effectively avoiding the forced increase of the gradient amplitude by the spike noise. Furthermore, while a single-scale sliding window can suppress isolated spike noise, it cannot account for the differences in latent heat release time among castings with different wall thicknesses. For example, thin-walled castings cool quickly and have short latent heat plateau durations. If the window is too long, the true solidification phase boundary will be over-smoothed and lost. Thick-walled castings cool slowly and have long latent heat plateau durations. If the window is too short, false secondary oscillations will occur when the window slides inside the plateau, interfering with phase boundary positioning. Therefore, the preset fixed window length is difficult to adapt to changes in casting wall thickness, leading to phase boundary extraction failure. Thus, this invention requires the preset of multiple windows of different scales for analysis.

[0027] In this embodiment of the invention, a short-scale window length N, a medium-scale window length 2N, and a long-scale window length 3N are preset. In this embodiment of the invention, the short-scale window length N is preset to 20. In other embodiments, the implementer may preset the value of N according to the specific implementation method. The higher quantile rank coordinate parameter H and the lower quantile rank coordinate parameter L are preset, wherein the value of H is between 0.5 and 1, and the value of L is between 0 and 0.5, and the sum of the higher quantile rank coordinate parameter H and the lower quantile rank coordinate parameter L is 1. One-dimensional linear sliding windows are established with the short-scale window length N, the medium-scale window length 2N, and the long-scale window length 3N respectively; the normalized temperature discrete sequence of any channel is denoted as the current normalized temperature discrete sequence; the one-dimensional linear sliding window of each scale slides forward along the current normalized temperature discrete sequence with a unit step size, and at each sampling point, all temperature sampling points in the window are rearranged in ascending order of value to obtain the local rank sequence of each sampling point at each scale; Obtain the normalized gradient magnitude at each sampling point at each scale: The product of the short-scale window length N and H is rounded up to obtain the high quantile index value. The value at the corresponding position of the high quantile index value is extracted from the local rank sequence of each sampling point under the short scale, and is used as the local maximum value of each sampling point under the short scale. The product of the short-scale window length N and the low quantile rank coordinate parameter L is rounded up to obtain the low quantile index value. The value at the corresponding position of the low quantile index value is extracted from the local rank sequence of each sampling point under the short scale, which is used as the local minimum value of each sampling point under the short scale. By discarding isolated sampling points at the extreme positions at both ends of the local rank sequence, abnormal extreme values ​​caused by electromagnetic interference or mechanical vibration are avoided from participating in the gradient calculation. The product of the short-scale window length N and 0.5 is rounded up to obtain the median index value. The value at the corresponding position of the median index value is extracted from the local rank sequence of each sampling point under the short scale, and used as the median rank reference value of each sampling point under the short scale. The median rank reference value is used to characterize the overall geometric center level of the temperature signal within the window. Obtain the normalized gradient magnitude for each sampling point at a short scale: In the formula, This represents the normalized gradient magnitude at the m-th sampling point under a short scale. This represents the local maximum value of the m-th sampling point at a short scale. This represents the local minimum value at the m-th sampling point under a short scale. This represents the median rank reference value of the m-th sampling point at a short scale. Representing a minimal constant, to avoid the denominator being zero, in this embodiment of the invention, a preset value is used. This normalization operation converts the absolute gradient magnitude into a relative slope index, eliminating the deviation between the absolute temperature level and the cooling rate magnitude caused by differences in the cooling stages, so that the gradient sequences of different solidification stages have a unified geometric scale benchmark. Finally, the normalized gradient magnitudes of all sampling points at the short scale are concatenated in chronological order of sliding time to form a short-scale morphological gradient sequence.

[0028] Similarly, based on the method of obtaining short-scale morphological gradient sequences, morphological gradient sequences at each scale are obtained.

[0029] S003. Based on the geometric confidence modulation factor of each sampling point in the current standardized temperature discrete sequence, obtain the multi-scale fusion morphological gradient magnitude of each sampling point; based on the flipping feature of the temperature change direction within the symmetrical local window of each sampling point, obtain the normalized oscillation of each sampling point; use the normalized oscillation to correct the multi-scale fusion morphological gradient magnitude to obtain the final gradient magnitude of each sampling point.

[0030] It should be noted that, given that a normalized gradient magnitude is calculated for each sampling point under each scale window, this invention first sums the gradient magnitudes of the same sampling point at each scale, which is then used as the basic fusion component for that sampling point. Next, an inter-scale consistency verification mechanism is introduced at each sampling point. Only when the dispersion index of the gradient magnitude of the same sampling point under multiple scale windows is small is the basic fusion component in a high-confidence state and it is retained; otherwise, it is determined that there is scale mismatch or spurious perturbation, and the basic fusion component is in a low-confidence state and is suppressed.

[0031] In this embodiment of the invention, the arithmetic mean of the normalized gradient magnitude of each sampling point at all scales is taken as the normalized gradient mean of each sampling point; the square deviation of the normalized gradient magnitude of each sampling point at each scale relative to the normalized gradient mean is calculated, and the square root of the arithmetic mean of the square deviation is obtained to obtain the dispersion index of each sampling point, which reflects the degree of dispersion of the values ​​between scales. Secondly, based on the dispersion index and gradient level of the current sampling point, the specific operation to obtain the geometric modulation factor of the current sampling point is as follows: In the formula, The geometric confidence modulation factor represents the current sampling point; This represents the dispersion index of the current sampling point; This represents the normalized gradient mean of the current sampling point; Representing a minimal constant, to avoid the denominator being zero, in this embodiment of the invention, a preset value is used. When the gradients at all three scales are highly consistent, the dispersion index approaches zero and the geometric modulation factor approaches 1, reflecting a high-confidence state of complete preservation. When the gradient directions at all three scales diverge or the values ​​are significantly discrete, the dispersion index increases sharply, and the modulation factor decreases rapidly and approaches 0, reflecting a state of spurious perturbation that is fully suppressed. When the mean of the normalized gradient and the dispersion index are both zero, numerical stability is ensured by the minimal division constant to prevent zeroing, and the modulation factor degenerates to 1, thus avoiding being misjudged as low-confidence due to the simultaneous zeroing of the three scales. Finally, the sum of the normalized gradient magnitudes of each sampling point at all scales is taken as the basic fusion component of each sampling point. The product of the basic fusion component of each sampling point and the geometric modulation factor of each sampling point is taken as the multi-scale fusion morphological gradient magnitude of each sampling point. The multi-scale fusion morphological gradient magnitudes of all sampling points are concatenated in traversal order to form a scale fusion morphological gradient sequence.

[0032] It should be noted that while the preceding steps effectively suppress isolated peaks and scale mismatch issues, during the cooling process of heavy structural components such as the water jacket of an automotive engine block, the local temperature fluctuates continuously near the true phase transition inflection point due to the influence of internal thermal convection and the pulsating impact of the external cooling medium. This fluctuation is not a single isolated peak, but rather occurs repeatedly over a period of time. Furthermore, this type of fluctuation repeatedly widens the temperature difference within the local window, leading to an abnormal rise in the multi-scale fused morphological gradient within the phase transition platform. This causes the system to mistakenly believe that the phase transition platform has ended prematurely. Since the amplitude of this pseudo-gradient rise may be close to the gradient recovery process at the end of the true platform, it is difficult to stably distinguish it based solely on the morphological gradient amplitude.

[0033] Therefore, by statistically analyzing the reversal characteristics of the temperature change direction in the symmetrical local window of each sampling point on the current standardized temperature discrete sequence, the normalized oscillation of the symmetrical local window of each sampling point is obtained. The larger the normalized oscillation, the more frequent the temperature rise and fall alternate within the window, indicating the presence of transient oscillation interference, and attenuation is applied to the multi-scale fused morphological gradient amplitude of the sampling point. The smaller the normalized oscillation, the more stable the temperature changes unidirectionally along the time axis within the window, and the multi-scale fused morphological gradient amplitude of the sampling point is preserved.

[0034] In this embodiment of the invention, for each sampling point on the current standardized temperature discrete sequence, M sampling points are taken before and after that point to form a symmetrical local window, such as... Figure 2 As shown; in this embodiment of the invention, the preset number of pixels M=60. In other embodiments, the implementer may preset the value of M according to the specific implementation method. The directional state sequence of the symmetrical local window of each sampling point is obtained as follows: Each sampling point in the symmetrical local window of each sampling point is traversed one by one. The temperature value of the current sampling point is subtracted from the temperature value of its subsequent adjacent sampling points to obtain the temperature difference value of the current sampling point. If the temperature difference value is greater than zero, the current sampling point is determined to be in a temperature rising state, and the directional state mark is assigned as +1. If the temperature difference value is less than zero, the current sampling point is determined to be in a temperature falling state, and the directional state mark is assigned as -1. If the temperature difference value is equal to zero, the current sampling point is determined to be in a temperature level state, and the directional state mark is assigned as 0. The directional state marks of all sampling points in the symmetrical local window of each sampling point are arranged in chronological order of traversal to form a directional state sequence. The normalized oscillation of the symmetrical local window at each sampling point is obtained as follows: For the direction state sequence of the symmetrical local window at each sampling point, the total number of times the direction state flag changes from +1 to -1 and from -1 to +1 in the sequence is counted, which is used as the flip count of the symmetrical local window at each sampling point; the flip count of the symmetrical local window at each sampling point is divided by the maximum value of the flip counts of the symmetrical local windows at all sampling points to obtain the normalized oscillation of each sampling point. Subtract the normalized oscillation value of each sampling point from 1 to obtain the retention coefficient of each sampling point. Multiply the multiscale fused morphological gradient magnitude of each sampling point by the retention coefficient of each sampling point to obtain the final gradient magnitude of each sampling point.

[0035] S004. Determine the actual duration of the crystallization platform for each channel based on the final gradient amplitude of each sampling point. Compare the duration of the channel with the slowest solidification with the standard crystallization process reference duration and adjust the mold cooling water flow rate accordingly.

[0036] It should be noted that by sequentially traversing each sampling point on the current standardized temperature discrete sequence, the duration of the actual crystallization platform of the alloy is extracted based on the final gradient amplitude of each sampling point, and compared with the process standard value. Based on this, the flow rate of the mold cooling water is automatically adjusted to achieve closed-loop control of the solidification process.

[0037] In this embodiment of the invention, firstly, the maximum value of the final gradient amplitude of all sampling points in the current standardized temperature discrete sequence is obtained as the global peak amplitude; in this embodiment of the invention, eight percent of the global peak amplitude is used as the entry threshold, and ten percent of the global peak amplitude is used as the exit threshold, wherein the value of the exit threshold is higher than the entry threshold. The number of consecutive sampling points is preset to K=10. The sampling points on the current standardized temperature discrete sequence are traversed sequentially. If the final gradient magnitude of the sampling point and its subsequent K sampling points are all less than the entry threshold, the time corresponding to the sampling point is the phase boundary start time. The traversal continues along the sampling point. If the final gradient magnitude of any sampling point and its subsequent K sampling points is greater than the exit threshold, the time corresponding to the sampling point is the phase boundary end time. The difference between the phase boundary end time stamp and the phase boundary start time stamp is taken as the actual crystallization platform duration of the current channel. Similarly, the actual duration of the solidification plateau in each channel is obtained, and the maximum value among the actual durations of the solidification plateau in each channel is taken as the final duration of the solidification plateau in this casting. It should be noted that the duration corresponding to the slowest solidification position is used for control.

[0038] Finally, the cooling deviation value is obtained by subtracting the final crystallization platform duration of the casting from the preset standard crystallization process reference duration. The cooling water circuit regulating valve of the mold is directly driven to adjust in the same direction according to the positive or negative direction of the cooling deviation value. When the cooling deviation value is positive (i.e., the actual crystallization time is longer than the reference time, and the solidification is slower), the cooling water flow rate is increased to accelerate heat dissipation. When the cooling deviation value is negative (i.e., the actual crystallization time is shorter than the reference time, and the solidification is faster), the cooling water flow rate is decreased to mitigate thermal shock. This completes the transformation from microscopic feature perception to macroscopic process closed-loop control.

[0039] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An optimization method for automotive casting forming process, characterized in that, include: Multi-channel temperature signals during the solidification process of castings are acquired, and after synchronization alignment, baseline compensation and geometric normalization mapping, a standardized temperature discrete sequence for each channel is obtained. For any channel's standardized temperature discrete sequence, a linear sliding window of multiple scales is used to slide through the sample points in unit step sizes. For each sample point, the local rank sequence of each sample point at each scale is obtained based on the temperature magnitude relationship of the sample points within the window. Based on the local rank sequence, the normalized gradient magnitude of each sample point at each scale is obtained. Based on the normalized gradient magnitude of the same sampling point at each scale, the geometric confidence modulation factor of each sampling point is obtained. The product of the geometric confidence modulation factor and the sum of the normalized gradient magnitudes of each sampling point at all scales is used as the multi-scale fused morphological gradient magnitude of each sampling point; based on the flipping characteristics of the temperature change direction within the symmetrical local window of each sampling point, the normalized oscillation of each sampling point is obtained. The normalized oscillations are used to correct the multi-scale fused morphological gradient magnitude to obtain the final gradient magnitude of each sampling point; The actual duration of the crystallization platform in each channel is determined based on the final gradient amplitude. The duration of the channel with the slowest solidification is compared with the standard crystallization process reference duration, and the flow rate of the mold cooling water is adjusted accordingly.

2. The method for optimizing the forming process of automotive castings according to claim 1, characterized in that, The process of obtaining the local rank sequence of each sampling point at each scale includes: The preset short-scale window length is N, the medium-scale window length is 2N, and the long-scale window length is 3N. For any channel's standardized temperature discrete sequence, a one-dimensional linear sliding window is established with the short-scale window length N, the medium-scale window length 2N, and the long-scale window length 3N. The one-dimensional linear sliding window at each scale is slid forward along the standardized temperature discrete sequence in unit steps. At each sampling point, all temperature sampling points within the window are rearranged in ascending order of value to obtain the local rank sequence of each sampling point at each scale.

3. The method for optimizing the forming process of automotive castings according to claim 1 or 2, characterized in that, The process of obtaining the normalized gradient magnitude of each sampling point at each scale includes: Obtain the local maximum, local minimum, and median rank reference values ​​for each sampling point at a short scale; In the formula, This represents the normalized gradient magnitude at the m-th sampling point under a short scale. This represents the local maximum value of the m-th sampling point at a short scale. This represents the local minimum value at the m-th sampling point under a short scale. This represents the median rank reference value of the m-th sampling point at a short scale. This represents a very small constant.

4. The method for optimizing the forming process of automotive castings according to claim 3, characterized in that, The acquisition of local maximum, local minimum, and median rank reference values ​​for each sampling point at a short scale includes: Preset the high quantile rank coordinate parameter H and the low quantile rank coordinate parameter L; calculate the product of the short-scale window length N with H, L and 0.5, and round up to obtain the high quantile index value, low quantile index value and median index value; take the value at the position corresponding to the high quantile index value in the local rank sequence of each sampling point under the short scale as the local maximum value of each sampling point under the short scale; take the value at the position corresponding to the low quantile index value in the local rank sequence of each sampling point under the short scale as the local minimum value of each sampling point under the short scale; take the value at the position corresponding to the median index value in the local rank sequence of each sampling point under the short scale as the median rank reference value of each sampling point under the short scale.

5. The method for optimizing the forming process of automotive castings according to claim 1, characterized in that, The step of obtaining the geometric confidence modulation factor for each sampling point includes: The arithmetic mean of the normalized gradient magnitudes of each sampling point at all scales is taken as the normalized gradient mean of each sampling point; the square of the deviation of the normalized gradient magnitude of each sampling point at each scale from the normalized gradient mean is calculated, and the square root of the arithmetic mean of the deviation squares is taken to obtain the dispersion index of each sampling point. In the formula, The geometric confidence modulation factor represents the current sampling point; This represents the dispersion index of the current sampling point; The normalized gradient mean of the current sampling point; This represents a very small constant.

6. The method for optimizing the forming process of automotive castings according to claim 1, characterized in that, The process of obtaining the normalized oscillation value at each sampling point includes: For each sampling point on the standardized temperature discrete sequence of any channel, take M sampling points in front of and behind that point to form a symmetrical local window; obtain the orientation state sequence of the symmetrical local window for each sampling point; For the direction state sequence of the symmetrical local window at each sampling point, the total number of times the direction state flag changes from +1 to -1 and from -1 to +1 in the sequence is counted, which is used as the flip count of the symmetrical local window at each sampling point; the flip count of the symmetrical local window at each sampling point is divided by the maximum value of the flip counts of the symmetrical local windows at all sampling points to obtain the normalized oscillation of the symmetrical local window at each sampling point.

7. The method for optimizing the forming process of automotive castings according to claim 1, characterized in that, The step of obtaining the directional state sequence of the symmetrical local window for each sampling point includes: The sampling points in the symmetrical local window of each sampling point are traversed one by one. The temperature value of the sampling point is subtracted from the temperature value of the subsequent adjacent sampling points of any sampling point to obtain the temperature difference value of the sampling point. If the temperature difference value is greater than zero, the sampling point is determined to be in a temperature rising state and is assigned a direction state mark of +1. If the temperature difference value is less than zero, the sampling point is determined to be in a temperature falling state and is assigned a direction state mark of -1. If the temperature difference value is equal to zero, the sampling point is determined to be in a temperature level state and is assigned a direction state mark of 0. The direction state marks of all sampling points in the symmetrical local window of each sampling point are arranged in chronological order of traversal to form a direction state sequence.

8. The method for optimizing the forming process of automotive castings according to claim 1, characterized in that, The step of correcting the multi-scale fused morphological gradient magnitude using the normalized oscillation to obtain the final gradient magnitude of each sampling point includes: Subtract the normalized oscillation value of each sampling point from 1 to obtain the retention coefficient of each sampling point. Multiply the multiscale fused morphological gradient magnitude of each sampling point by the retention coefficient of each sampling point to obtain the final gradient magnitude of each sampling point.

9. The method for optimizing the forming process of automotive castings according to claim 1, characterized in that, The determination of the actual crystallization platform duration for each channel includes: The maximum value of the final gradient amplitude of all sampling points in the current standardized temperature discrete sequence is taken as the global peak amplitude. Eight percent of the global peak amplitude is taken as the entry threshold, and ten percent of the global peak amplitude is taken as the exit threshold. A preset number of continuous sampling points K is used to traverse each sampling point in the current standardized temperature discrete sequence. If the final gradient amplitude of the sampling point and its subsequent K sampling points are all less than the entry threshold, the time corresponding to the sampling point is the phase boundary start time. Continue to traverse along the sampling point. If the final gradient amplitude of any sampling point and its subsequent K sampling points is greater than the exit threshold, the time corresponding to the sampling point is taken as the phase boundary end time. The difference between the phase boundary end timestamp and the phase boundary start timestamp is taken as the actual crystallization platform duration of the current channel.

10. The method for optimizing the forming process of automotive castings according to claim 1, characterized in that, The process of comparing the time corresponding to the slowest solidification channel with the standard crystallization process reference time and adjusting the mold cooling water flow rate includes: The maximum value among the actual crystallization plateau durations of each channel is taken as the final crystallization plateau duration of this casting. The cooling deviation value is obtained by subtracting the final crystallization plateau duration of this casting from the preset standard crystallization process reference duration. When the cooling deviation value is positive, the cooling water flow rate is increased to accelerate heat dissipation. When the cooling deviation value is negative, the cooling water flow rate is decreased to mitigate thermal shock.

Images