Intelligent evaluation method and system for heat treatment quality of castings

Abstract

This invention discloses an intelligent evaluation method and system for the heat treatment quality of castings, relating to the field of quality evaluation technology. The method includes constructing a process state vector and automatically segmenting the heating, holding, and cooling stages using a strategy combining CUSUM cumulative values ​​and process window constraints. Based on a set of thermal deviation points, it constructs a triple feature system encompassing multi-scale occupancy structural complexity, local neighborhood growth complexity, and local distribution deviation. An adversarial deletion mechanism is used to extract and fuse deviation values. Combining temperature deviation and cumulative anomaly values, the method calculates the heat treatment quality evolution evaluation value. This invention constructs a process state vector synchronized with multi-source data and automatically identifies stage switching points based on incremental sequence CUSUM. Combined with multi-scale spatial structural complexity analysis, it achieves precise segmentation of heat treatment stages and fine local anomaly localization, improving the real-time performance and accuracy of casting heat treatment quality evaluation.

Description

Technical Field

[0001] This invention relates to the field of quality assessment technology, and in particular to an intelligent assessment method and system for the quality of heat treatment of castings. Background Technology

[0002] As a key process for controlling material properties, heat treatment of castings mainly relies on manual experience and offline testing methods. With the development of industrial internet and process monitoring technology, multi-source sensors are widely deployed in heat treatment equipment, realizing continuous monitoring of the furnace environment. By introducing statistical analysis methods and process control theory, the state identification and anomaly detection of the heat treatment process are carried out, enabling the quality assessment of heat treatment to evolve from a single indicator to a multi-dimensional state characterization.

[0003] However, existing technologies have shortcomings. Current heat treatment sites use absolute temperature thresholds or fixed time points for stage division and quality assessment. This method is simple and easy to implement, but because it does not combine the synchronous changes of multi-source state vectors, nor does it use incremental features such as temperature rise rate and cooling capacity, it cannot accurately reflect the actual process stage switching and dynamic changes of the thermal field. Once the equipment operating conditions fluctuate or there are batch differences, misjudgments or judgment lags are likely to occur, making it difficult to ensure both the feasibility of the project and the accuracy of the assessment. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent evaluation method and system for the heat treatment quality of castings, which solves the problem of using absolute temperature thresholds or fixed time nodes for stage division and quality evaluation in existing heat treatment sites. This method is simple to implement and easy to implement, but because it does not combine the synchronous changes of multi-source state vectors, nor does it use incremental features such as temperature rise rate and cooling capacity, it cannot accurately reflect the actual process stage switching and dynamic changes of the thermal field. Once the equipment operating conditions fluctuate or there are differences between batches, misjudgments or judgment lags are likely to occur, making it difficult to ensure both the feasibility of the project and the accuracy of the evaluation.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an intelligent evaluation method for the heat treatment quality of castings, which includes the following steps:

[0008] The heat treatment monitoring unit is acquired, a process state vector is generated, the temperature rise increment and cooling intensity increment are calculated, and the candidate intervals for temperature rise and cooling boundaries are defined in combination with the process formula. Within the interval, the candidate points for the end of temperature rise and the candidate points for the start of cooling are screened by the cumulative amount of CUSUM, and the automatic stage segmentation result set is set.

[0009] Based on the automatic stage segmentation result set, the furnace body measurement points are divided into partitions, a thermal deviation point set is constructed and normalized, the final structural complexity deviation of the normalized thermal deviation point set is extracted, the local outlier value at a single moment of the partition is calculated, the cumulative outlier of the partition is constructed, the dominant outlier region and the peak cumulative outlier are located, the heat treatment quality evolution evaluation value is generated and the threshold is determined, and the evaluation result is output.

[0010] As a preferred embodiment of the intelligent evaluation method for casting heat treatment quality described in this invention, the step of partitioning the furnace body measuring points, constructing a set of thermal deviation points and performing normalization processing, and extracting the final structural complexity deviation of the normalized thermal deviation point set includes:

[0011] All measuring points of the furnace body are fixedly partitioned according to the furnace body structure. The horizontal coordinates, vertical coordinates, and normalized temperature deviation height of the measuring points in each partition are extracted and sorted horizontally to obtain a set of thermal deviation points. The set of thermal deviation points is then normalized to generate a normalized set of thermal deviation points.

[0012] Extract the multi-scale occupied structural complexity deviation, local neighborhood growth complexity deviation, and local distribution deviation over-boundary of the normalized thermal deviation point set;

[0013] Each point in the normalized thermal deviation point set is deleted sequentially to obtain the deleted point set. The multi-scale occupancy structure complexity, local neighborhood growth complexity, and local distribution deviation of the deleted points are extracted, and the instability penalty amount is obtained through threshold judgment.

[0014] The final structural complexity deviation is generated by fusing the multi-scale occupation structural complexity deviation, the local neighborhood growth complexity deviation, the local distribution deviation from the limit, and the instability penalty.

[0015] As a preferred embodiment of the intelligent evaluation method for casting heat treatment quality described in this invention, the calculation of local anomalies at a single moment in a partition, the construction of cumulative anomaly quantities in the partition, and the location of the dominant anomaly region and the peak cumulative anomaly quantity include:

[0016] Calculate and average the representative temperature deviation, temperature dispersion deviation, and structural complexity deviation to obtain the local outlier value at a single moment. Construct the cumulative outlier. Sort the cumulative outliers of all partitions in descending order. Select the partition with the largest cumulative outlier and set it as the anomaly-dominant region. Select the largest cumulative outlier within the anomaly-dominant region and set it as the peak cumulative outlier.

[0017] As a preferred embodiment of the intelligent evaluation method for casting heat treatment quality described in this invention, the step of generating heat treatment quality evolution evaluation values, determining thresholds, and outputting evaluation results includes:

[0018] The stage length is defined as the end time of the stage minus the start time of the stage. The heat treatment quality evolution evaluation value is calculated and a threshold is used for judgment to obtain quality stability, quality warning and quality risk.

[0019] As a preferred embodiment of the intelligent evaluation method for casting heat treatment quality described in this invention, the set of automatically segmented results includes:

[0020] Define heating boundary verification quantities and cooling boundary verification quantities, filter candidate points for the end of heating and the start of cooling to obtain valid boundary points for the end of heating and the start of cooling, and set the automatic stage segmentation result set.

[0021] As a preferred embodiment of the intelligent evaluation method for casting heat treatment quality described in this invention, the step of screening candidate points for the end of heating and the start of cooling by accumulating CUSUM within a given interval includes:

[0022] Calculate the cumulative CUSUM within the candidate intervals for heating boundary and cooling boundary, and sort them in descending order. Select the candidate point with the largest cumulative CUSUM as the candidate point for the end of heating and the candidate point for the start of cooling.

[0023] As a preferred embodiment of the intelligent evaluation method for casting heat treatment quality described in this invention, the step of acquiring the heat treatment monitoring unit, generating a process state vector, calculating the temperature rise increment and cooling intensity increment, and defining candidate intervals for temperature rise and cooling boundaries in conjunction with the process formula includes:

[0024] The monitoring units of heat treatment are collected through the API interface, a process state vector is constructed, the temperature rise increment and cooling intensity increment are calculated, and two candidate intervals are set in combination with the process formula, including the temperature rise boundary candidate interval and the cooling boundary candidate interval.

[0025] Secondly, the present invention provides an intelligent evaluation system for the heat treatment quality of castings, comprising:

[0026] The data acquisition module is used to acquire monitoring units at time k through the API interface, including furnace temperature, furnace pressure, furnace oxygen content, cooling status, and measuring point location;

[0027] The state generation module is used to extract representative temperature and temperature dispersion, calculate cooling intensity, and generate corresponding process state vectors.

[0028] The variable point screening module is used to calculate the temperature rise increment and cooling intensity increment based on the process state vector, define the boundary candidate interval in combination with the process formula, and screen two types of stage candidate points through the CUSUM cumulative amount.

[0029] The boundary verification module is used to define two types of boundary verification quantities, and combine the verification window and the judgment threshold to complete the validity verification of candidate points.

[0030] The beneficial effects of this invention are as follows: This invention automatically segments the heating, holding, and cooling stages by constructing a process state vector and employing a strategy combining CUSUM cumulative quantity and process window constraints. Based on the thermal deviation point set, it constructs a triple feature of multi-scale occupied structural complexity, local neighborhood growth complexity, and local distribution deviation. Through an adversarial deletion mechanism, it extracts and merges deviation quantities. Combining temperature deviation and cumulative anomaly quantity, it calculates the heat treatment quality evolution evaluation value. This invention constructs a process state vector with multi-source data synchronization and automatically identifies stage switching points based on incremental sequence CUSUM. Combined with multi-scale spatial structural complexity analysis, it achieves accurate segmentation of heat treatment stages and fine local anomaly location, improving the real-time performance and accuracy of casting heat treatment quality assessment. Attached Figure Description

[0031] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart of the intelligent evaluation method for the heat treatment quality of castings in Example 1.

[0033] Figure 2 This is a schematic diagram of the intelligent evaluation system for casting heat treatment quality in Example 2. Detailed Implementation

[0034] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0035] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides an intelligent evaluation method for the heat treatment quality of castings, including the following steps:

[0036] S1. Obtain the heat treatment monitoring unit, generate the process state vector, calculate the temperature rise increment and cooling intensity increment, and define the candidate intervals for temperature rise and cooling boundaries in combination with the process formula. Within the interval, screen the candidate points for the end of temperature rise and the candidate points for the start of cooling by the cumulative amount of CUSUM, and set the automatic stage segmentation result set.

[0037] Specifically, the heat treatment monitoring unit acquires data, generates a process state vector, calculates the temperature rise increment and cooling intensity increment, and, in conjunction with the process formulation, delineates candidate intervals for temperature rise and cooling boundaries, including:

[0038] Collecting time via API interface The monitoring unit includes furnace temperature, furnace pressure, furnace oxygen content, cooling status, and measuring point locations;

[0039] The cooling state includes the cooling medium flow rate and the cooling medium temperature.

[0040] The location of the measuring point refers to the fixed spatial location of the temperature measuring point inside the furnace.

[0041] It can provide homogeneous, synchronous, and directly computable data inputs for subsequent automatic segmentation and local anomaly localization, avoiding misjudgments caused by inconsistent timing of different data streams;

[0042] It should be noted that all historical data used in subsequent steps are obtained synchronously through the API interface in this step;

[0043] The arithmetic mean method is used to extract the average value of the temperature values ​​in the monitoring unit to obtain the time. The represented temperature;

[0044] The arithmetic mean method was chosen to extract the representative temperature because one of the purposes of multi-point temperature measurement in the heat treatment furnace is to obtain the current overall thermal level. The arithmetic mean method has a clear source, is simple to implement, can be directly reproduced in the engineering field, and is easy to correlate with the process target temperature. It compresses the multi-point temperature into an overall thermal state quantity that can be used for process tracking, and provides a stable benchmark for subsequent temperature uniformity characterization and stage change detection.

[0045] The temperature dispersion was calculated using the standard deviation method for both the temperature values ​​and the representative temperatures.

[0046] The standard deviation method is chosen because the heat treatment quality of castings is affected not only by the overall temperature level but also by the temperature difference in the furnace space. The standard deviation can directly reflect the degree of dispersion of the temperature at multiple points relative to the overall average temperature, which is suitable for describing the uniformity of the thermal field. This allows the process state vector to contain both the overall thermal level and the degree of uniformity of the thermal field, thereby avoiding the risk of ignoring local temperature differences when judging solely by the average temperature.

[0047] Based on the cooling status of the monitoring unit, the sum of the cooling medium temperature value and the denominator correction is calculated, and then the cooling medium flow rate value is divided by the sum to obtain the cooling intensity.

[0048] The denominator correction amount is used to avoid denominator abnormalities when the cooling medium temperature is close to zero, and is set to a fixed value of 0.001;

[0049] Under the same equipment conditions, the greater the flow rate, the stronger the cooling medium's ability to remove heat, and the lower the medium temperature, the higher the cooling potential. By placing the flow rate in the numerator and the medium temperature in the denominator, a monotonic index can be formed that increases with the flow rate and decreases with the medium temperature. This makes it convenient to use as the core state variable in the cooling stage, so that the final state vector can cover the changes in cooling capacity during the heat treatment process, and provide direct input for the identification of the cooling stage and the analysis of cooling deviations.

[0050] The parameters representing temperature, temperature dispersion, oxygen content, furnace pressure, and cooling intensity are sorted in a fixed order, and the generation time is determined accordingly. The process state vector (as a column vector);

[0051] Based on the process state vector, using time... The representative temperature minus the time The representative temperature is used to obtain the temperature increase increment;

[0052] The reason for choosing to represent the temperature increment rather than the absolute temperature value as the detection object is that the stage boundary essentially reflects the switching of the state change law. The most essential difference between the heating stage and the heat preservation stage is not the magnitude of the temperature value, but whether the temperature continues to rise.

[0053] Based on the process state vector, using time... Cooling intensity minus time The cooling intensity is calculated, and the cooling intensity increment is obtained.

[0054] The reason for choosing the incremental cooling intensity rather than the absolute cooling intensity is that the beginning of the cooling phase is manifested as a sudden change in the cooling effect from weak to strong, and this change is more easily and accurately identified in the incremental sequence than in the absolute sequence.

[0055] Based on the process formulation, two candidate intervals are defined, including the candidate interval for the heating boundary. and cooling boundary candidate interval ;

[0056] The process formula refers to a set of standardized process parameters that are pre-defined and stored in the control system for the target casting material, model, size and heat treatment process requirements. It is used to guide the heat treatment equipment to complete the entire process of heating, holding and cooling according to the preset trajectory.

[0057] The , , as well as The start time number, end time number, start time number, and end time number of cooling boundary monitoring are obtained by dividing the earliest end time of heating, the latest end time of heating, the earliest start time of cooling, and the latest start time of cooling in the process formula by the sampling period, respectively.

[0058] The candidate detection range is introduced because although the switching of heat treatment stages is determined by the actual process, it is still constrained by the process formula and cannot occur arbitrarily at any time. By using the process context to limit the detection window, interference from local fluctuations unrelated to the process can be eliminated, the boundary search range can be narrowed, false change points can be reduced, and the stability and feasibility of stage switching identification can be improved.

[0059] Furthermore, within the interval, candidate points for the end of heating and the start of cooling are screened using the cumulative CUSUM, including:

[0060] Calculate the average value of the temperature rise increment within the candidate interval of the temperature rise boundary, using the time interval... The cumulative CUSUM plus the average, then subtracting the time interval. The temperature increase is obtained at time The cumulative CUSUM values ​​are sorted in descending order, and the maximum cumulative CUSUM value is selected as the candidate point for the end of the heating process.

[0061] The reason for choosing to accumulate positive CUSUM for the decrease in temperature rise increment is that the essential manifestation of the end of the temperature rise phase is that the continuous temperature rise trend weakens and stabilizes into a near-zero change state. CUSUM is most sensitive to this kind of mean drift switching and can automatically extract the true position of the temperature rise stop from the continuous temperature rise process, avoiding boundary shift caused by judging based on fixed duration or a single threshold.

[0062] Repeat the above operation for the cooling intensity increment to obtain candidate cooling start points;

[0063] The reason for choosing to accumulate positive CUSUM for the increase in cooling intensity is that the core characteristic of the beginning of the cooling stage is not an immediate drop in temperature, but rather the activation and continuous enhancement of cooling capacity. This is the most direct in the cooling intensity increment sequence, which can accurately identify the true starting point of the cooling stage and avoid misjudging the natural temperature fluctuations at the end of the insulation as the start of cooling.

[0064] Furthermore, the automatic stage segmentation result set is defined, including:

[0065] Define heating boundary verification quantities and cooling boundary verification quantities, filter candidate points for heating end and cooling start, obtain valid boundary points for heating end and cooling start, and set the automatic stage segmentation result set, using the following formula:

[0066] ,

[0067] ,

[0068] in, This is the temperature rise boundary verification value for the back window. The cooling boundary verification value has the same formula as the temperature rise boundary verification value. To verify the window length, To verify the offset sequence number within the window, For the increase in temperature, To end the candidate point of warming, This is the set of results from automatic stage segmentation. This is the sequence number of the last sampling time in the current batch. The warming phase is underway. For the heat preservation stage, This is the cooling phase;

[0069] The verification window length refers to the time of use. timestamp minus time The timestamp is used to obtain the sampling period. The shortest stable duration after the stage switch (obtained from the process formula) is divided by the sampling period and rounded up to obtain the verification window length.

[0070] The process of filtering candidate points includes calculating the temperature rise boundary verification value of the front window, dividing the temperature rise boundary verification value of the back window by the sum of the temperature rise boundary verification value and the correction value of the front window to obtain the temperature decay determination ratio, and selecting candidate points for the end of temperature rise that have a temperature decay determination ratio less than the temperature decay determination threshold and meet the consistency verification as valid boundary points for the end of temperature rise. Calculate the cooling boundary check value of the front window, divide the cooling boundary check value of the back window by the sum of the absolute value of the cooling boundary check value of the front window and the correction value, and obtain the cooling enhancement judgment ratio. Select cooling start candidate points whose cooling enhancement judgment ratio is greater than the cooling enhancement judgment threshold and meet the consistency check, and set them as valid cooling start boundary points. ;

[0071] The temperature decay judgment threshold is set based on the statistical analysis of historical qualified batch data. After the temperature rise ends, the temperature rise increment decreases significantly, and the mean value of the later window is much smaller than that of the earlier window. It can be set to 0.3.

[0072] The cooling enhancement judgment threshold is set based on the statistical analysis of historical qualified batch data. After cooling starts, the cooling intensity increases significantly and the average value of the later window is significantly greater than that of the earlier window. It can be set to 2.

[0073] The front window is the candidate point for the end of heating minus the length of the verification window to the candidate point for the end of heating minus 1, and the back window is the candidate point for the end of heating plus the length of the verification window to the candidate point for the end of heating minus 1. The window structure for the candidate point for the start of cooling is the same.

[0074] The consistency verification of the candidate point for the end of heating refers to calculating the mean of temperature dispersion, mean of oxygen content, and mean of furnace pressure of the front and rear windows. The conditions are met as follows: the mean of temperature dispersion of the rear window is less than or equal to the sum of the mean of temperature dispersion of the front window and the allowable temperature dispersion fluctuation bandwidth of the initial heat preservation stage; the absolute value of the mean of oxygen content of the rear window minus the mean of oxygen content of the front window is less than or equal to the allowable oxygen content fluctuation bandwidth when heating to heat preservation; and the absolute value of the mean of furnace pressure of the rear window minus the mean of furnace pressure of the front window is less than or equal to the allowable furnace pressure fluctuation bandwidth when heating to heat preservation.

[0075] The consistency check of the candidate cooling start point refers to the average temperature rise increment of the calculated window and the difference between the maximum temperature dispersion and the minimum temperature dispersion of the calculated window. The conditions are that the average temperature rise increment is less than 0 and the difference is less than or equal to the upper limit of the allowable temperature dispersion range in the initial cooling stage, and the screening conditions of oxygen content and furnace pressure are met.

[0076] It should be noted that the above allowable values ​​were all obtained through process formulation;

[0077] The reason for choosing CUSUM variable point detection instead of fixed time segmentation is that CUSUM is most sensitive to the boundary between one stable statistical law and another. It is suitable for identifying the stage switching points in heat treatment, such as the transition from continuous heating to steady-state heat preservation and from steady-state heat preservation to forced cooling. It can automatically eliminate the statistical aliasing introduced by the stage switching transition segment, so that the heating, heat preservation and cooling stages each maintain a single physical semantic in the subsequent evaluation, thereby improving the evaluation accuracy.

[0078] S2. Based on the automatic stage segmentation result set, the furnace body measurement points are divided into partitions, a thermal deviation point set is constructed and normalized, the final structural complexity deviation of the normalized thermal deviation point set is extracted, the local outlier value at a single moment of the partition is calculated, the cumulative outlier of the partition is constructed, the dominant outlier area and the peak cumulative outlier are located, the heat treatment quality evolution evaluation value is generated and the threshold is determined, and the evaluation result is output.

[0079] Specifically, the furnace body measurement points are partitioned, a thermal deviation point set is constructed and normalized, and the final structural complexity deviation of the normalized thermal deviation point set is extracted, including:

[0080] For each stage in the automatic stage segmentation result set Obtain the target representative temperature and temperature tolerance bandwidth in the process formulation;

[0081] All measuring points on the furnace body are fixedly divided into zones according to the furnace structure, such as the front zone, middle zone, and rear zone, and zone numbers are assigned. ;

[0082] For the stage Middle partition The moment For each measuring point in the partition, the normalized temperature deviation height is obtained by subtracting the target temperature value from the temperature value and then dividing by the sum of the temperature tolerance bandwidth and correction amount, thus generating a three-dimensional deviation point. The horizontal coordinates, vertical coordinates, and normalized temperature deviation height of the corresponding measurement points within the partition are used to horizontally sort the three-dimensional deviation points of the same stage, the same partition, and the same time to obtain the thermal deviation point set.

[0083] The center point is calculated using the arithmetic mean method for all points in the hot deviation point set, and the feature spacing is extracted using the median method for the distance between each point in the hot deviation point set and its corresponding nearest neighbor.

[0084] The nearest neighbor refers to the Euclidean distance between each point and other points, obtained by sorting in ascending order.

[0085] The normalized points are obtained by subtracting the center point from the points in the hot deviation point set and then dividing by the sum of the feature spacing and the correction amount. This generates the normalized hot deviation point set.

[0086] It can upgrade temperature difference data, which originally only reflects single-point error, into spatial structure data, providing a direct carrier for subsequent multi-scale structural complexity extraction, thereby improving the ability to identify local thermal field instability.

[0087] The thermal deviation point set will have overall translation and amplitude differences in different furnace batches, zones and times. Only by performing center alignment and scale normalization can we ensure that the subsequent extraction is the change in thermal field structure, rather than positional shift or overall temperature difference increase. This can eliminate the interference of slight deviation in furnace loading position, overall drift of measuring points and absolute deviation amplitude differences on structural analysis, so that the subsequent complexity results can truly reflect the degree of local thermal field instability.

[0088] The minimum bounding cube side length of the normalized thermal deviation point set is defined as the structural span. A recursive relationship is defined between the number of scale layers and the number of axial divisions to obtain the voxel side length and the number of non-empty voxels, as shown in the formula:

[0089] ,

[0090] in, The number of scale layers, For scale layer numbering, It is a set of positive integers. For the first The number of divisions along each coordinate axis when dividing a layer into voxels. The number of scale layers allowed in the normalized thermal deviation point set, wherein the total number of voxels in the finest layer does not exceed the number of points in the normalized thermal deviation point set. For stage numbering, For partition numbering, For a specific moment;

[0091] The recursive relationship of the axial division number is as follows: , No. The side length of the layer voxel is the structural span divided by Statistics The number of non-empty voxels is obtained by normalizing the number of voxels occupied by the points in the thermal deviation point set. The voxel index of the point in the thermal deviation point set is obtained by subtracting the minimum value of the minimum bounding cube in the three directions from the three-dimensional coordinates of the point, dividing by the voxel side length, and then rounding down to obtain the voxel index, thus realizing the mapping from continuous coordinates to discrete voxel index.

[0092] The minimum circumscribed cube refers to the maximum and minimum values ​​of the three directions extracted from the normalized thermal deviation point set, the maximum value minus the minimum value to obtain the span, and the maximum span is set as the cube side length.

[0093] The natural logarithm of the voxel side length and the natural logarithm of the number of non-empty voxels are set as inputs. The least squares linear method is used to fit the input to obtain the multi-scale occupied structural complexity. ;

[0094] Anomalies in casting heat treatment often manifest as the expansion, agglomeration, or fragmentation of thermal deviation points in space. These are difficult to characterize using only a single-point temperature difference. Therefore, multi-scale occupancy structural complexity is used to describe the spatial spreading pattern of thermal deviation point sets. This can identify spatial anomaly patterns such as local thermal agglomeration, strip-shaped thermal shift, and dispersed thermal drift, thereby improving the ability to characterize the uneven changes in the thermal field organization within the furnace.

[0095] The neighborhood order is defined by the formula:

[0096] ,

[0097] in, Let the order be the neighborhood order. The number of points in the normalized thermal deviation point set. To ensure a lower bound on the minimum number of neighborhood points required for local dimension estimation, such as setting it to 10 or 20, and to satisfy statistical stability, the value should be more than twice the dimensionless estimate.

[0098] For any point E in the normalized set of thermal deviation points, the local intrinsic dimension of the point is calculated using the maximum likelihood estimation method based on the neighborhood order, and the local neighborhood growth complexity is extracted using the median method. ;

[0099] Multi-scale structures tend to focus on the overall morphology and are difficult to reveal the expansion rate within local neighborhoods. Therefore, further extraction of local neighborhood growth complexity is used to characterize the intrinsic growth characteristics of local thermal deviations from the structure. This can identify early thermal anomalies such as local cracking, local agglomeration, and local diffusion earlier, so that heat treatment quality assessment can not only look at the overall structure but also capture local instability trends.

[0100] Calculate the Euclidean distance from point E to the center of the normalized thermal deviation point set and sort them in descending order. Filter for the largest Euclidean distance and define the radial space. The formula is:

[0101] ,

[0102] ,

[0103] in, The number of radial shell layers. For the maximum Euclidean distance, The feature distance is obtained by using the median method to calculate the difference in Euclidean distance between all adjacent sorted pairs. For correction amount, For radial space, Number the shell layers;

[0104] The center of the normalized thermal deviation point set refers to the average value point of the normalized thermal deviation point set;

[0105] The radial distribution is obtained by counting the number of points in the normalized thermal deviation point set that are located in the radial interval and dividing by the number of points in the normalized thermal deviation point set.

[0106] For the historical thermal deviation point set (referring to the set of points formed by merging thermal deviation points from the same stage and zone of all heat-treated batches with the same material, model, and process formula as the current batch, and whose quality is deemed qualified), extract the historical radial distribution (using the same method as described above). Then, use Jensen-Shannon divergence to extract the local distribution deviation from both the historical radial distribution and the radial distribution. ;

[0107] Abnormal heat treatment not only manifests as increased structural complexity, but may also manifest as the deviation point migrating from the center to the edge or from the edge to the center. Therefore, it is necessary to extract the local distribution deviation to describe the radial distribution change, which can identify the migration direction and distribution center of gravity of the thermal deviation structure, avoid missing the thermal field state with the same complexity but abnormal distribution pattern, and improve the completeness of the assessment.

[0108] Each point in the normalized thermal deviation point set is deleted sequentially. Each deletion results in a set of deleted points that are missing one point.

[0109] Repeat the above operation on the set of points to be deleted to obtain the structural complexity occupied by the multi-scale deletion. Local neighborhood growth complexity and local distribution deviation ;

[0110] Individual thermocouple drift, transient interference, or sampling glitch may form pseudo-complex structures. Therefore, it is necessary to verify whether the current structure still has stability after deletion by single-point countermeasure deletion. This can effectively suppress false alarms caused by single-point anomalies, make the system pay more attention to real thermal anomalies with group characteristics, persistence, and spatial organization, and improve the reliability of the assessment.

[0111] The 95th percentiles of the multi-scale occupancy structural complexity, local neighborhood growth complexity, and local distribution deviation of historical qualified batches are extracted to obtain the multi-scale occupancy structural complexity threshold. Local neighborhood growth complexity threshold and the local distribution deviation threshold ;

[0112] If deleting the point satisfies , and If the deletion point is positive, mark it as 1; otherwise, mark it as 0. Sum and average the deletion point marks to obtain the single-point anti-deletion degree.

[0113] use Divide by The sum of the corrections yields the multi-scale occupancy structural complexity deviation, which is then used... Divide by The sum of the corrections yields the deviation of the local neighborhood growth complexity, which is then used. Divide by The sum of the corrections yields the amount by which the local distribution deviates from the limit.

[0114] Subtract the single-point adversarial deletion degree from 1 to obtain the unstable penalty amount;

[0115] The multi-scale structural complexity deviation, local neighborhood growth complexity deviation, local distribution deviation from the limit, and instability penalty are horizontally sorted to obtain the fused component set. The absolute value of the number of elements in the fused component set is set as the number of components. The final structural complexity deviation is then calculated using the following formula:

[0116] ,

[0117] in, This represents the deviation in final structural complexity. For consecutive multiplication, For the fusion component set, The number of components, For each deviation component in the fused component set;

[0118] Occupation structure, neighborhood growth, distribution migration, and deletion stability reflect different aspects of thermal deviation from the structure. Using any one of these indicators alone may be one-sided. Therefore, unweighted fusion is required to form a unified deviation value. The result has both the ability to identify spatial structural anomalies and the ability to suppress pseudo-structures. It can be directly used as input for subsequent zonal cumulative anomaly analysis, thereby improving the accuracy and stability of intelligent assessment of heat treatment quality.

[0119] Furthermore, the local outliers at a single time point in the partition are calculated, the cumulative outlier of the partition is constructed, and the dominant outlier region and peak cumulative outlier are located, including:

[0120] First calculation stage Middle partition The moment The absolute value of the target representative temperature is subtracted from the representative temperature, and then the absolute value is divided by the sum of the temperature allowable deviation bandwidth and the correction amount to obtain the representative temperature deviation. The above steps are repeated for the temperature dispersion to obtain the temperature dispersion deviation.

[0121] The structural complexity deviation is obtained by dividing the final structural complexity deviation by the sum of the maximum value of the reference index (the maximum value of the final structural complexity deviation in historical qualified batches) and the correction amount.

[0122] The local outliers at a single time moment are obtained by summing and averaging the temperature deviation, temperature dispersion deviation, and structural complexity deviation, and the cumulative outlier is constructed using the following formula:

[0123] ,

[0124] in, To accumulate outliers, This represents a local outlier at a single moment.

[0125] Sort the cumulative outliers of all partitions in descending order, select the partition with the largest cumulative outlier as the outlier-dominant region, and select the largest cumulative outlier within the outlier-dominant region as the peak cumulative outlier.

[0126] This step uses a method of accumulating abnormal values ​​by zone and locating the dominant area, rather than scoring the entire furnace individually. This is because heat treatment defects often first appear in local areas and only spread to the whole after continuous accumulation in the local area. This method can not only indicate whether the stage is abnormal, but also indicate which area (front, middle, or rear) is abnormal first, which is convenient for direct guidance on targeted adjustments to the blower, furnace loading, furnace door sealing, spraying, or stirring systems.

[0127] Furthermore, it generates heat treatment quality evolution evaluation values ​​and determines thresholds, outputting evaluation results, including:

[0128] The stage length is defined as the difference between the stage's end time and its start time. The heat treatment quality evolution evaluation value is calculated using the following formula:

[0129] ,

[0130] in, This is a value used to evaluate the evolution of heat treatment quality. The length of the heating phase. The length of the insulation stage, This refers to the length of the cooling stage. This refers to the peak cumulative abnormal amount during the heating phase. This represents the peak cumulative abnormal amount during the insulation stage. This represents the peak cumulative abnormality during the cooling phase.

[0131] Set early warning thresholds, such as 1.5 times the standard deviation or 80th percentile of the quality evolution evaluation value of historical qualified batches; set risk thresholds, such as 95th percentile of the quality evolution evaluation value of historical qualified batches.

[0132] If the heat treatment quality evolution evaluation value is less than the warning threshold, it is judged as stable in quality; if the heat treatment quality evolution evaluation value is greater than or equal to the warning threshold but less than the risk threshold, it is judged as a quality warning; if the heat treatment quality evolution evaluation value is greater than or equal to the warning threshold, it is judged as a quality risk.

[0133] It can unify automatic stage segmentation, local structural complexity, and partition cumulative anomalies into a final quality conclusion, while retaining the most critical stage and region location results, which can be directly used to serve on-site control.

[0134] Example 2, refer to Figure 2 As a second embodiment of the present invention, an intelligent evaluation system for the heat treatment quality of castings includes:

[0135] The data acquisition module is used to acquire monitoring units at time k through the API interface, including furnace temperature, furnace pressure, furnace oxygen content, cooling status, and measuring point location;

[0136] The state generation module is used to extract representative temperature and temperature dispersion, calculate cooling intensity, and generate corresponding process state vectors.

[0137] The variable point screening module is used to calculate the temperature rise increment and cooling intensity increment based on the process state vector, define the boundary candidate interval in combination with the process formula, and screen two types of stage candidate points through the CUSUM cumulative amount.

[0138] The boundary verification module is used to define two types of boundary verification quantities, and combine the verification window and the judgment threshold to complete the validity verification of candidate points.

[0139] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent evaluation of heat treatment quality of a casting, characterized in that: Includes the following steps: The heat treatment monitoring unit is acquired, a process state vector is generated, the temperature rise increment and cooling intensity increment are calculated, and the candidate intervals for temperature rise and cooling boundaries are defined in combination with the process formula. Within the interval, the candidate points for the end of temperature rise and the candidate points for the start of cooling are screened by the cumulative amount of CUSUM, and the automatic stage segmentation result set is set. Based on the automatic stage segmentation result set, the furnace body measurement points are divided into partitions, a thermal deviation point set is constructed and normalized, the final structural complexity deviation of the normalized thermal deviation point set is extracted, the local outlier value at a single moment of the partition is calculated, the cumulative outlier of the partition is constructed, the dominant outlier region and the peak cumulative outlier are located, the heat treatment quality evolution evaluation value is generated and the threshold is determined, and the evaluation result is output.

2. The method of claim 1, wherein: The process of partitioning the furnace body measuring points, constructing a thermal deviation point set, normalizing it, and extracting the final structural complexity deviation of the normalized thermal deviation point set includes: All measuring points of the furnace body are fixedly partitioned according to the furnace body structure. The horizontal coordinates, vertical coordinates, and normalized temperature deviation height of the measuring points in each partition are extracted and sorted horizontally to obtain a set of thermal deviation points. The set of thermal deviation points is then normalized to generate a normalized set of thermal deviation points. Extract the multi-scale occupied structural complexity deviation, local neighborhood growth complexity deviation, and local distribution deviation over-boundary of the normalized thermal deviation point set; Each point in the normalized thermal deviation point set is deleted sequentially to obtain the deleted point set. The multi-scale occupancy structure complexity, local neighborhood growth complexity, and local distribution deviation of the deleted points are extracted, and the instability penalty amount is obtained through threshold judgment. The final structural complexity deviation is generated by fusing the multi-scale occupation structural complexity deviation, the local neighborhood growth complexity deviation, the local distribution deviation from the limit, and the instability penalty.

3. The method of claim 2, wherein: The calculation of local outliers in a single time period within a partition, the construction of cumulative outlier values ​​for the partition, and the location of the dominant outlier region and peak cumulative outlier values ​​include: Calculate and average the representative temperature deviation, temperature dispersion deviation, and structural complexity deviation to obtain the local outlier value at a single moment. Construct the cumulative outlier. Sort the cumulative outliers of all partitions in descending order. Select the partition with the largest cumulative outlier and set it as the anomaly-dominant region. Select the largest cumulative outlier within the anomaly-dominant region and set it as the peak cumulative outlier.

4. The method of claim 3, wherein: The process of generating heat treatment quality evolution evaluation values, determining thresholds, and outputting evaluation results includes: The stage length is defined as the end time of the stage minus the start time of the stage. The heat treatment quality evolution evaluation value is calculated and a threshold is used for judgment to obtain quality stability, quality warning and quality risk.

5. The intelligent evaluation method for the heat treatment quality of castings as described in claim 4, characterized in that: The set of automatic stage segmentation results includes: Define heating boundary verification quantities and cooling boundary verification quantities, filter candidate points for the end of heating and the start of cooling to obtain valid boundary points for the end of heating and the start of cooling, and set the automatic stage segmentation result set.

6. The intelligent evaluation method for the heat treatment quality of castings as described in claim 5, characterized in that: The process of screening candidate points for the end of heating and the start of cooling within the interval using the cumulative CUSUM value includes: Calculate the cumulative CUSUM within the candidate intervals for heating boundary and cooling boundary, and sort them in descending order. Select the candidate point with the largest cumulative CUSUM as the candidate point for the end of heating and the candidate point for the start of cooling.

7. The intelligent evaluation method for the heat treatment quality of castings as described in claim 1, characterized in that: The heat treatment monitoring unit generates a process state vector, calculates the temperature rise increment and cooling intensity increment, and, in conjunction with the process formula, delineates candidate intervals for temperature rise and cooling boundaries, including: The monitoring units of heat treatment are collected through the API interface, a process state vector is constructed, the temperature rise increment and cooling intensity increment are calculated, and two candidate intervals are set in combination with the process formula, including the temperature rise boundary candidate interval and the cooling boundary candidate interval.

8. A smart evaluation system for the heat treatment quality of castings, used to implement the smart evaluation method for the heat treatment quality of castings as described in any one of claims 1 to 7, characterized in that: include: The data acquisition module is used to acquire monitoring units at time k through the API interface, including furnace temperature, furnace pressure, furnace oxygen content, cooling status, and measuring point location; The state generation module is used to extract representative temperature and temperature dispersion, calculate cooling intensity, and generate corresponding process state vectors. The variable point screening module is used to calculate the temperature rise increment and cooling intensity increment based on the process state vector, define the boundary candidate interval in combination with the process formula, and screen two types of stage candidate points through the CUSUM cumulative amount. The boundary verification module is used to define two types of boundary verification quantities, and combine the verification window and the judgment threshold to complete the validity verification of candidate points.

Images