Method and device for evaluating the degradation state of the sealing performance of an integrated circuit high-temperature aging station

Abstract

The application discloses a kind of integrated circuit high-temperature aging station sealing performance degradation state evaluation method and device, belong to integrated circuit technical field, this method includes: to the temperature variation curve of each temperature sensor corresponding temperature with time variation is carried out feature extraction, obtains temperature curve feature sequence;According to temperature curve feature sequence determines the first temperature distribution vector of each position and leakage parameter estimate value;Each temperature feature sequence is input into the physical drive model established in advance, predicts the second temperature distribution vector of each position;According to second temperature distribution vector and the actual measured temperature distribution vector of each position collected, generate temperature distance consistency evaluation index;According to temperature distribution vector and temperature distance consistency evaluation index, obtain target state vector;According to target state vector, determine high-temperature aging station test chamber sealing health index;According to sealing health index and the leakage coefficient estimate value at each position, determine high-temperature aging station test chamber door plate sealing performance.

Description

Technical Field

[0001] This invention relates to the field of integrated circuit technology, and in particular to a method and apparatus for evaluating the degradation status of sealing performance on a high-temperature aging test bench for integrated circuits. Background Technology

[0002] The high-temperature test chamber on the high-temperature aging bench controls the internal temperature to test the performance and reliability of electronic devices, materials, and other products under high-temperature conditions. During the high-temperature test, the uniformity of temperature distribution and the stability of temperature control inside the chamber directly affect the accuracy and repeatability of the test results. The sealing performance of the test chamber, as a crucial factor in ensuring the isolation of the internal temperature environment from the external environment, plays a vital role in the effectiveness of the high-temperature test.

[0003] As the test chamber ages, problems such as aging door sealing strips, loose door hinges, and deformation of the inner structure of the oven door gradually emerge, leading to a gradual deterioration of the chamber's sealing performance. Under high-temperature operating conditions, this decreased sealing performance causes heat leakage to the outside of the chamber, especially near the door panel, resulting in significant temperature field deviations and localized abnormal temperature rises. This not only affects the overall temperature uniformity within the test chamber but may also cause some tested products to be placed in environments that do not meet testing specifications, leading to distorted test data or even test failure.

[0004] It is evident that there is an urgent need for those skilled in the art to provide a solution that can timely and reliably assess the degradation status of the sealing performance of a high-temperature aging test bench. Summary of the Invention

[0005] The purpose of this invention is to provide a method and apparatus for evaluating the degradation status of sealing performance on a high-temperature aging test bench for integrated circuits, which can solve at least one of the above-mentioned problems in the prior art.

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

[0007] This invention provides a method for evaluating the degradation status of sealing performance on a high-temperature aging test bench for integrated circuits, wherein the method includes:

[0008] The temperature information of each temperature sensor deployed in a preset area around the door panel of the high-temperature aging test chamber is collected, and a temperature change curve over time is generated.

[0009] Feature extraction is performed on the temperature change curves of each temperature sensor over time to obtain the feature sequence of the temperature rise curves corresponding to each temperature sensor.

[0010] The characteristic sequences of each heating curve and the historical temperature sequences of each temperature sensor are input into a pre-established data-driven model to obtain the first temperature distribution vector and leakage parameter estimate for each location. The first temperature distribution vector contains the predicted temperature value for each location at the next moment. Each temperature sensor corresponds to one location.

[0011] Each of the temperature feature sequences is input into a pre-established physical driving model to predict the second temperature distribution vector at each of the locations;

[0012] Based on the second temperature distribution vector and the actual measured temperature distribution vectors collected at each of the aforementioned locations, a temperature distance consistency evaluation index is generated;

[0013] Based on the first temperature distribution vector, the second temperature distribution vector, and the temperature distance consistency evaluation index, the target state vector is obtained;

[0014] The target state vector is recursively estimated to obtain the leakage coefficient estimate at each of the locations;

[0015] The sealing health index of the high-temperature aging test chamber is calculated based on the estimated leakage coefficient at each of the aforementioned locations.

[0016] The sealing performance of the high-temperature aging test chamber door is determined based on the sealing health index and the estimated leakage coefficient at each of the aforementioned locations.

[0017] This invention provides a device for evaluating the sealing performance degradation status of an integrated circuit high-temperature aging test bench, the device comprising:

[0018] The first generation module is used to collect temperature information from various temperature sensors arranged in a preset area around the door panel of the high-temperature aging test chamber and generate a temperature change curve over time.

[0019] The feature extraction module is used to extract features from the temperature change curves of each temperature sensor over time to obtain the feature sequence of the heating curves of each temperature sensor.

[0020] The first estimation module is used to input the characteristic sequence of each of the heating curves and the historical temperature sequence of each of the temperature sensors into a pre-established data-driven model to obtain the first temperature distribution vector and the leakage parameter estimate for each location. The first temperature distribution vector contains the temperature prediction value for each location at the next moment. Each temperature sensor corresponds to one location.

[0021] The prediction module is used to input the temperature feature sequences into a pre-established physical driving model to predict the second temperature distribution vector at each location;

[0022] The second generation module is used to generate a temperature distance consistency evaluation index based on the second temperature distribution vector and the actual measured temperature distribution vectors collected at each of the locations.

[0023] The target state vector determination module is used to obtain the target state vector based on the first temperature distribution vector, the second temperature distribution vector, and the temperature distance consistency evaluation index.

[0024] The second estimation module is used to recursively estimate the target state vector to obtain the leakage coefficient estimate at each of the locations.

[0025] The calculation module is used to calculate the sealing health index of the high-temperature aging bench test chamber based on the estimated leakage coefficient at each of the aforementioned locations.

[0026] The determination module is used to determine the sealing performance of the high-temperature aging test chamber door panel based on the sealing health index and the estimated leakage coefficient at each of the locations.

[0027] This invention provides an electronic device, which includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the steps of any of the above-described methods for evaluating the sealing performance degradation status of integrated circuits in a high-temperature aging test bench.

[0028] This invention provides a readable storage medium storing a program or instructions, which, when executed by a processor, implement the steps of any of the above-described methods for evaluating the sealing performance degradation of an integrated circuit high-temperature aging test bench.

[0029] The integrated circuit high-temperature aging test chamber sealing performance degradation assessment scheme provided in this embodiment of the invention collects temperature information from various temperature sensors arranged in a preset area around the door panel of the high-temperature aging test chamber, generating temperature-time variation curves; extracts features from the temperature-time variation curves corresponding to each temperature sensor to obtain the temperature rise curve feature sequence corresponding to each temperature sensor; inputs the temperature rise curve feature sequence and the historical temperature sequence of each temperature sensor into a pre-established data-driven model to obtain the first temperature distribution vector and leakage parameter estimate for each location; inputs the temperature feature sequence into a pre-established physical-driven model to predict the second temperature distribution vector for each location; generates a temperature distance consistency evaluation index based on the second temperature distribution vector and the actual measured temperature distribution vector collected at each location; obtains the target state vector based on the first temperature distribution vector, the second temperature distribution vector, and the temperature distance consistency evaluation index; and recursively estimates the target state vector to obtain the leakage coefficient estimate for each location.

[0030] The sealing health index of the high-temperature aging test chamber is calculated based on the estimated leakage coefficients at various locations. The sealing performance of the high-temperature aging test chamber door is then determined based on the sealing health index and the estimated leakage coefficients at various locations. The solution provided in this embodiment of the invention, firstly, integrates the advantages of both physical-driven and data-driven models. The physical-driven model provides physical interpretability and stability, while the data-driven model provides strong data fitting capabilities. Their complementarity makes the model both accurate and generalizable. Secondly, the nonlinear modeling capability is enhanced. The data-driven model can capture complex nonlinear relationships in the temperature rise curve, making the system more sensitive to heat leakage at door gaps, especially under small sample conditions. Thirdly, by monitoring the distribution of the leakage coefficient along distance, spatial location of local sealing anomalies can be achieved. Using a consistency-weighted extended Kalman filter algorithm, physical prediction and data-driven prediction are dynamically fused. With temperature-distance consistency as the confidence factor, the observation weights are adaptively adjusted, enabling high-precision, real-time prediction of sealing performance parameters. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the steps of a method for evaluating the sealing performance degradation status of an integrated circuit high-temperature aging test bench according to an embodiment of this application.

[0032] Figure 2 This is a flowchart illustrating the steps of a method for evaluating the sealing performance degradation status of an integrated circuit high-temperature aging test bench according to an embodiment of this application.

[0033] Figure 3 This is a flowchart illustrating the steps of a method for assessing the health of the heating wire in a high-temperature test chamber for aging integrated circuits, according to an embodiment of this application.

[0034] Figure 4 This is a flowchart illustrating the steps of a health assessment method for a fan in a high-temperature test chamber for an integrated circuit aging bench, according to an embodiment of this application.

[0035] Figure 5 This is a structural block diagram illustrating an integrated circuit high-temperature aging test bench sealing performance degradation status assessment device according to an embodiment of this application. Detailed Implementation

[0036] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0037] This invention provides a scheme for evaluating the degradation state of sealing performance in a high-temperature aging chamber for integrated circuits. Firstly, the scheme collects temperature rise curves as a function of time and distance by deploying a one-dimensional or two-dimensional temperature sensor array near the door panel, establishing a physical driving model based on the heat diffusion mechanism and the characteristics of the chamber. Secondly, it extracts short-term features using a temporal convolutional network (TCN) and enhances nonlinear fitting capabilities by combining it with a Kolmogorov-Arnold network (KAN). Thirdly, it innovatively proposes a consistency-weighted extended Kalman filter (C-EKF) algorithm to dynamically fuse physical prediction and data-driven prediction, using temperature-distance consistency as a confidence factor to adaptively adjust observation weights, achieving high-precision, real-time estimation of sealing performance parameters. This scheme not only maintains stability under conditions of partial sensor failure or noise interference but also identifies local sealing anomalies through consistency indices, thereby achieving the goal of inferring the sealing health status from temperature field information.

[0038] The following description, in conjunction with the accompanying drawings, details the integrated circuit high-temperature aging bench sealing performance degradation assessment scheme provided in this application through specific embodiments and application scenarios.

[0039] As attached Figure 1 As shown, the method for evaluating the sealing performance degradation status of an integrated circuit high-temperature aging test bench according to an embodiment of this application includes the following steps:

[0040] Step 101: Collect temperature information from various temperature sensors located within a preset area around the door panel of the high-temperature aging test chamber, and generate a temperature change curve over time.

[0041] The integrated circuit high-temperature aging bench sealing performance degradation status assessment method provided in this embodiment of the invention is applied to an electronic device. The electronic device includes a processor and a memory. The memory stores an integrated circuit high-temperature aging bench sealing performance degradation status assessment program or instructions. When the integrated circuit high-temperature aging bench sealing performance degradation status assessment program or instructions are executed by the processor, the integrated circuit high-temperature aging bench sealing performance degradation status assessment method flow is realized.

[0042] In practical implementation, it can be installed near the door panel of the incubator. Each temperature sensor is equidistant, and the temperature change curve of each sensor over time is recorded:

[0043]

[0044] in, For time, The distance from the door panel.

[0045] By processing the raw data collected by each temperature sensor through filtering, noise reduction, normalization, synchronous interpolation, and anomaly detection, a set of time-series matrix data can be obtained.

[0046] When filtering and denoising, moving average or Savitzky-Golay filtering can be used.

[0047] Normalization can be performed by interval normalization, and the normalization formula is as follows: ;

[0048] Synchronous interpolation ensures that the time axes of all temperature sensors are consistent; anomaly detection can eliminate signals with obvious drift or dead zones.

[0049] A set of time series matrix data can be obtained. ,in This represents the number of sampling times.

[0050] Step 102: Extract features from the temperature change curves of each temperature sensor over time to obtain the feature sequence of the temperature rise curves corresponding to each temperature sensor.

[0051] In one optional embodiment, the method of extracting features from the temperature change curves of each temperature sensor over time to obtain the heating curve feature sequence of each temperature sensor can be as follows: for the temperature change curve of each temperature sensor over time, extract the initial heating rate, equivalent time constant, temperature curve area and spatial temperature gradient features to generate the heating curve feature sequence of the temperature sensor.

[0052] In practical implementation, the temperature change curve over time at each temperature sensor location can be obtained in the following way. Extract representative features:

[0053] 1) Initial temperature rise rate

[0054]

[0055] in: This is a fixed reference time after the heating process of the incubator begins, usually selected as a short time point in the early stage of heating; This indicates the initial response rate of temperature at that location.

[0056] This feature reflects the rate of heat transfer from the interior of the enclosure to the door panel area and is highly sensitive to localized heat leakage caused by a decline in sealing performance. These characteristics reflect the effects of heat diffusion rate, localized heat capacity effect, and sealing condition on the temperature field distribution.

[0057] 2) Equivalent time constant

[0058] By fitting the temperature rise curve with an exponential model, we obtain:

[0059]

[0060] in: This is the steady-state temperature at that location; This is the equivalent time constant for the temperature rise process.

[0061] time constant The rate at which the temperature at this location reaches a steady state is characterized by the combined effects of local heat capacity, thermal resistance, and the door panel's sealing condition. When the sealing performance deteriorates, the temperature near the door panel... Significant changes usually occur.

[0062] 3) Area of ​​the temperature curve

[0063]

[0064] in: This is the end time of the selected analysis time window.

[0065] This feature represents the accumulated heat level at that location within a given time interval, and is used to comprehensively reflect the impact of long-term heat retention or continuous leakage on the temperature response.

[0066] 4) Spatial temperature gradient

[0067]

[0068] in: and Spatial location of adjacent sensors. Spatial temperature gradient. This describes the temperature distribution along the normal direction of the door panel. When the door panel is poorly sealed, the temperature gradient near the leakage area will increase significantly.

[0069] Step 103: Input the characteristic sequence of each heating curve and the historical temperature sequence of each temperature sensor into the pre-established data-driven model to obtain the first temperature distribution vector and leakage parameter estimate for each location.

[0070] The first temperature distribution vector contains the predicted temperature value for each location at the next moment; each temperature sensor corresponds to one location.

[0071] After extracting the features of the temperature rise curve, in order to further characterize the dynamic law of temperature change with time and spatial location and capture its nonlinear characteristics, this application embodiment constructs a joint data-driven model based on the temporal convolutional network TCN and the Kolmogorov-Arnold network (KAN network). Based on this data-driven model, the feature sequence of the temperature rise curve corresponding to each temperature sensor is predicted, and the estimated value of the equivalent leakage coefficient at each distance position and the health status evaluation result of the sealing performance of the temperature chamber door panel calculated by it are output.

[0072] In an optional embodiment, inputting the characteristic sequences of each heating curve and the historical temperature sequences of each temperature sensor into a pre-established data-driven model to obtain the predicted temperature value and leakage parameter estimate for each location at the next moment may include the following sub-steps:

[0073] Sub-step 1: Based on the characteristic sequence of each heating curve and the historical temperature sequence of each temperature sensor, generate the first time-series feature vector corresponding to each location.

[0074] This step is the process of constructing the input data for the data-driven model.

[0075] For the distance from the door panel is The sensor at the location, at discrete time Construct the timing input vector:

[0076]

[0077] in: For the past at this location Temperature sequence at each sampling time; The temperature rise curve features extracted in step 102 are used. The above inputs simultaneously include dynamic time information and statistical features with clear physical meaning.

[0078] Sub-step 2: Input each first temporal feature vector into the TCN module of the data-driven model for convolution operation to obtain the second temporal feature vector corresponding to each position.

[0079] The TCN module performs convolution operations on the input first temporal feature vector to extract dynamic features of temperature changes. Its single-layer dilated convolution expression is as follows:

[0080]

[0081] in: Expansion rate; The kernel size; For the first The temporal characteristics of each moment. For the first Each convolutional kernel weight matrix is ​​used to perform linear mapping and weighted combination of input features at different time lag positions; This is the bias vector, used to adjust the baseline level of the convolution output;

[0082] The residual structure of the TCN module ensures gradient stability in long-term series modeling, enabling it to effectively capture the dynamic evolution trend of the temperature rise process.

[0083] Sub-step 3: Input each second time-series feature vector into the KAN module of the data-driven model for nonlinear mapping to obtain the temperature prediction value of each location at the next time step.

[0084] This step involves the nonlinear mapping enhancement process of the KAN module. To enhance the model's ability to represent complex nonlinear thermal processes, the second time-series feature vector output from the TCN module is input into the KAN module for nonlinear mapping. Specifically, this can be achieved using the following formula:

[0085]

[0086] in: It is a learnable kernel function, represented by a combination of polynomials or Bessel basis functions; These are the corresponding weight parameters.

[0087] The KAN module learns activation functions to achieve high-order nonlinear approximation in a low-dimensional space, making it suitable for modeling thermal processes under small sample conditions.

[0088] In the aforementioned steps, the characteristics of the temperature rise curve and the historical temperature sequence have been constructed into the model input vector. That is, the first temporal feature vector, which is then processed sequentially by the TCN module and the KAN module.

[0089] The TCN module extracts dynamic features representing temperature changes over time from the input sequence, while the KAN module performs nonlinear mapping and compression on the temporal features in the time-series feature vector output by the TCN module. The two modules are cascaded to form a unified data-driven model.

[0090]

[0091] Based on the above data-driven model For the location located at a distance of the door panel The temperature sensor at the location, at any time Predict the temperature value at the next moment:

[0092]

[0093] in: It is composed of the historical temperature sequence and the temperature rise curve features extracted in step 102, that is, the feature sequence of each temperature rise curve and the historical temperature sequence of each temperature sensor are jointly constituted. This indicates the temperature prediction results obtained based on the data-driven model.

[0094] Sub-step 4: For each location, determine the estimated leakage parameters at the location based on the predicted temperature value at the next moment and the actual measured temperature value at the location.

[0095] Since changes in the sealing performance of the door panel directly affect the temperature evolution rate and spatial distribution characteristics, the temperature prediction results implicitly contain information related to the sealing status. Therefore, a leakage parameter estimation function is further constructed. Based on the relationship between the predicted temperature and the measured temperature at the previous moment, the equivalent leakage coefficient at the corresponding location is estimated:

[0096]

[0097] in: Implemented using regression networks or function mapping; The prior estimates of the leakage parameters, i.e., the slope parameter estimates, are obtained through data-driven methods.

[0098] At this point, the data-driven model has completed its output, including: the estimated equivalent leakage coefficient at each distance location and the health status evaluation results of the chamber door sealing performance calculated from it.

[0099] Step 104: Input each temperature feature sequence into the pre-established physical driving model to predict the second temperature distribution vector at each location.

[0100] This step involves predicting the temperature distribution inside the test chamber based on a physical model of thermal diffusion, also known as a physics-driven model. In this step, a one-dimensional thermal diffusion model is established along the normal distance direction of the door panel, based on the heat conduction mechanism inside the test chamber (also called a temperature chamber), to describe the ideal evolution of the temperature field under given sealed conditions.

[0101]

[0102] in: Indicates time The distance from the door panel is The temperature inside the chamber; Represents a continuous-time variable; This represents the spatial coordinates along the normal direction of the door panel, i.e., the distance from the door panel; The thermal diffusivity is used to characterize the ability of heat to diffuse along the distance direction inside the box. The equivalent leakage coefficient, which varies with spatial location, is used to characterize the intensity of the effect of door panel sealing performance on heat loss. Indicates the external ambient temperature of the test chamber; This is the heating source item, used to represent the heat input of the heating system inside the chamber.

[0103] By performing finite difference discretization on the above partial differential equation in time and space, we obtain the discrete evolution equation of temperature:

[0104]

[0105] in: For discrete-time index, indicating the first... Each sampling time; For the first Each spatial discrete position represents a distance from the door panel. Location; The time interval between walks corresponds to the temperature sampling time interval. The spatial distance between the walk distance corresponds to the distance between adjacent temperature sensors; For the first Time, location Temperature at that location; The temperature at the next moment is predicted by the physics-driven model; For position The equivalent leakage coefficient at the location.

[0106] This discrete model is used to predict a physically consistent temperature distribution that satisfies the thermal diffusion mechanism, given the assumption of a leakage coefficient distribution:

[0107]

[0108] This result is the second temperature distribution vector, which serves as an ideal physical reference for subsequent consistency evaluation.

[0109] Step 105: Based on the second temperature distribution vector and the actual measured temperature distribution vectors collected at each location, generate a temperature distance consistency evaluation index.

[0110] Because the degradation of door panel sealing performance disrupts the normal heat diffusion along the distance direction, the actual measured temperature distribution will deviate from the ideal temperature distribution predicted by the physical model. To quantitatively characterize the degree of this deviation, a temperature-distance consistency index C is introduced. k It is used to evaluate how well the current actual temperature field follows the laws of heat diffusion. This is used to reflect the degree of agreement between the currently measured temperature distribution and the physically consistent temperature distribution. When the door panel's sealing performance deteriorates and heat leakage intensifies, the temperature field will deviate from the normal heat diffusion law, thus leading to... Significant decrease.

[0111] In one optional embodiment, the method for generating a temperature distance consistency evaluation index based on a first temperature distribution vector and the actual measured temperature distribution vectors collected at each location may include the following sub-steps:

[0112] Sub-step 1: Calculate the gradient consistency index, shape similarity index, and energy ratio consistency index between the first temperature distribution vector and the actual measured temperature distribution vectors collected at each location;

[0113] Sub-step 2: Weight the gradient consistency index, shape similarity index, and energy ratio consistency index to generate the temperature distance consistency evaluation index.

[0114] In practical implementation, temperature distance consistency evaluation index C k The definition can be as follows:

[0115]

[0116] in: For the first The overall temperature distance consistency index at each moment; This is a gradient consistency index;

[0117] It is a shape similarity index; As an indicator of energy ratio consistency; These are the weighting coefficients for each indicator, i.e., the consistency component, used to adjust the relative importance of each consistency component in the overall evaluation, and satisfying the following conditions: .

[0118] An exemplary gradient consistency The formula for calculating can be as follows:

[0119]

[0120] in: This represents the total number of temperature sensors in the distance direction; For actual temperature measurement at location Spatial gradient at a given location; Predict the spatial gradient of temperature at corresponding locations for the physics-driven model.

[0121] Gradient consistency measures whether the rate of temperature change along the distance direction matches the predictions of a thermal diffusion model. When local leakage intensifies, the temperature gradient will change abruptly, causing this indicator to drop significantly.

[0122] An exemplary shape similarity The formula for calculating can be as follows:

[0123]

[0124] in: For the first The actual measured temperature distribution vector along the distance direction at any given moment; The temperature distribution vector predicted by the physical driving model is the second temperature distribution vector. The correlation coefficient is used to measure the similarity between two distributions in terms of their overall shape.

[0125] The shape similarity index reflects whether the overall spatial distribution of the temperature field is consistent with the prediction results of the physical model.

[0126] An exemplary energy ratio consistency The formula for calculating can be as follows:

[0127]

[0128] in: The energy used to actually measure the temperature distribution; Predict the energy of temperature distribution for a physics-driven model.

[0129] Energy ratio consistency measures the relative relationship between the overall temperature level and the predictions of the physical model, and can be used to identify overall heat loss due to seal leakage.

[0130] Step 106: Based on the first temperature distribution vector, the second temperature distribution vector, and the temperature distance consistency evaluation index, obtain the target state vector.

[0131] In an optional embodiment, the target state vector can be obtained based on the first temperature distribution vector, the second temperature distribution vector, and the temperature distance consistency evaluation index by: calculating the observation noise based on the temperature distance consistency evaluation index; and summing the first temperature distribution vector, the second temperature distribution vector, and the observation noise to obtain the target state vector.

[0132] This step improves the Extended Kalman Filter (C-EKF) fusion. Specifically, to achieve stable and continuous estimation of leakage parameters under conditions of measurement noise and model uncertainty, this embodiment introduces a weighting mechanism based on the temperature-distance consistency evaluation index on the basis of the traditional Extended Kalman Filter (EKF) framework to fuse the prediction results of the physics-driven model and the output of the data-driven model. When improving the Extended Kalman Filter (EKF), the following operations are performed:

[0133] 1) Definition of state variables

[0134] Define the system state vector as follows:

[0135]

[0136] in: For the corresponding distance position The equivalent leakage coefficient at the location; The number of discrete spatial points along the distance direction; This is the ambient temperature offset, used to compensate for changes in the external environment and measurement deviations.

[0137] 2) State transition model

[0138] Considering the slow-changing nature of the sealing performance degradation process and the limited change in the leakage coefficient between adjacent time points, the state transition model is expressed as:

[0139]

[0140] in, This is process noise, used to characterize seal aging and random disturbance factors.

[0141] 3) Construction of observation model

[0142] The system observations are the actual temperature measurements at each measuring point, and their relationship with the state variables is described by the following two parts:

[0143] Physically predicted temperature calculated by the physics-driven model ;

[0144] Temperature prediction values ​​obtained from data-driven models .

[0145] Combining the two types of prediction information mentioned above, the observation equation can be expressed as:

[0146]

[0147] The observation equation is the improved extended Kalman filter C-EKF fusion formula, where: This is the actual measured temperature vector; To observe noise.

[0148] This application also provides a consistency-based adaptive adjustment scheme for observation noise, as detailed below:

[0149] To reflect the degree to which the current temperature field conforms to the physical laws of heat transfer, a temperature distance consistency index is introduced. And based on this, dynamically adjust the observation noise covariance matrix:

[0150]

[0151]

[0152] When temperature distance consistency index When the temperature is low, it indicates that the actual temperature distribution deviates from the ideal heat diffusion law. In this case, increasing the observation noise reduces the impact of the observation data on the state estimation; when... When the value is high, the reliability of the observation is improved.

[0153] Step 107: Perform recursive estimation on the target state vector to obtain the leakage coefficient estimate at each location.

[0154] This step involves filtering, updating, and outputting the result. Specifically, based on the extended Kalman filter prediction and update formula, the target state vector is recursively estimated, ultimately yielding:

[0155]

[0156] in, Indicates the first Leakage coefficient estimates at each location.

[0157] Step 108: Calculate the sealing health index of the high-temperature aging test chamber based on the estimated leakage coefficient values ​​at each location.

[0158] Steps 108-109 are based on the leakage coefficient estimation results obtained from C-EKF fusion, and are used to conduct health assessments and anomaly determinations on the overall and local sealing performance of the test chamber door.

[0159] In the specific implementation process, the overall sealing health index of the test chamber can be defined as:

[0160]

[0161] in: For a moment Overall sealing health index; The first one estimated by C-EKF Leakage coefficient at each location; This is the sensitivity adjustment coefficient.

[0162] The sealing health index decreases monotonically as the degree of leakage increases, and is used to characterize the overall degradation level of the door panel sealing performance.

[0163] Step 109: Determine the sealing performance of the high-temperature aging test chamber door panel based on the sealing health index and the estimated leakage coefficient at each location.

[0164] In one optional embodiment, the sealing performance of the high-temperature aging test chamber door can be determined based on the sealing health index and the estimated leakage coefficient at each location as follows:

[0165] In the sealed health index H kLess than the first preset threshold H th Under these circumstances, it was determined that the overall sealing performance of the high-temperature aging test chamber door panel had deteriorated;

[0166] For each location, estimate the leakage coefficient at that location. With the second preset threshold k th By comparing and filtering, the target locations of local sealing abnormalities can be identified.

[0167] If any location meets the following condition, it is determined that there is a local sealing abnormality at that location:

[0168]

[0169] In practical implementation, the estimated leakage coefficient at each location can be compared with the second preset threshold using the above formula to identify target locations with local sealing anomalies. Precisely pinpointing the sealing anomaly to its local location facilitates more accurate and targeted anomaly handling.

[0170] In an optional embodiment, the integrated circuit high-temperature aging test bench sealing performance degradation assessment method provided in this application further includes an early deterioration warning mechanism, which can be implemented as follows:

[0171] A time series trend analysis was performed on the temperature distance consistency evaluation index to obtain the trend change results. When the trend change results indicate a continuous downward trend, it was determined that the door panel of the high temperature aging bench test chamber was in early deterioration, and an early warning was issued.

[0172] In one feasible implementation method, the early deterioration warning mechanism is set up so that when performing time series trend analysis on the temperature-distance consistency evaluation index, its derivative with time can be calculated. When it shows a continuous downward trend, i.e., the derivative is less than 0, it is determined that there are signs of early deterioration in the sealing performance, and an early warning is triggered. The specific implementation formula can be as follows:

[0173]

[0174] The integrated circuit high-temperature aging test bench sealing performance degradation assessment method provided in this application has the following advantages: First, it integrates the advantages of both physical-driven and data-driven models. The physical-driven model provides physical interpretability and stability, while the data-driven model provides strong data fitting capabilities. The complementarity of the two makes the model both accurate and generalizable. Second, the nonlinear modeling capability is enhanced. The data-driven model can capture complex nonlinear relationships in the temperature rise curve, making the system more sensitive to heat leakage at the door gap, especially under small sample conditions. Third, by monitoring the distribution of the leakage coefficient along the distance, spatial location of local sealing anomalies can be achieved. By using the consistency weighted extended Kalman filter algorithm, the physical prediction and data-driven prediction are dynamically fused. The observation weights are adaptively adjusted using temperature-distance consistency as the confidence factor, which can achieve high-precision, real-time prediction of sealing performance parameters.

[0175] The following reference Figure 2 The process of the integrated circuit high-temperature aging test bench sealing performance degradation assessment method provided in the embodiments of this application is summarized.

[0176] Temperature rise curves from various temperature sensors near the door panel of the high-temperature test chamber are collected. These curves reflect the temperature-distance distribution data. The collected temperature rise curves undergo data preprocessing and feature extraction to obtain a feature sequence. This feature sequence is then input into both the physical-driven and data-driven models for prediction. Based on the predicted values, a temperature-distance consistency metric is calculated, i.e., a temperature-distance consistency evaluation index. The outputs of the temperature-distance consistency evaluation index, the physical-driven model, and the data-driven model are fused using C-EKF to obtain a comprehensive health index, i.e., the test chamber's sealing health index, and an estimated leakage coefficient distribution. This allows for the identification of localized sealing degradation locations based on the leakage coefficient distribution.

[0177] The integrated circuit high-temperature aging stage sealing performance degradation assessment method provided in this application introduces a consistency-weighted extended Kalman filter (C-EKF) mechanism: utilizing the temperature gradient along the distance direction, the correlation between adjacent points, and the overall distribution similarity, a temperature-distance consistency evaluation index is constructed to achieve health confidence assessment based on spatial characteristics. Furthermore, the temperature-distance consistency evaluation index is incorporated into the dynamic adjustment of the filter observation covariance matrix to achieve adaptive weight allocation based on physical consistency, improving robustness under local anomalies or noisy environments.

[0178] Furthermore, a multi-source information fusion model is introduced, integrating both physics-driven and data-driven models. The data-driven model incorporates TCN and KAN networks, replacing traditional fully connected layers with learnable activation functions to enhance the nonlinear mapping ability between feature layers and improve the model's generalization performance under small sample sizes or complex thermal environments. The physics-driven model incorporates heat conduction and leakage observation models. By fusing the temperature rise predictions from both the physics-driven and data-driven models, the system possesses both physical interpretability and strong nonlinear fitting capabilities.

[0179] This application also provides a health assessment scheme for the heating wire and the fan of a high-temperature test chamber for integrated circuit aging. The integrated circuit high-temperature aging chamber sealing performance degradation assessment scheme can combine the above two schemes to assess the health status of the aging chamber. The specific logic for combining the assessment results of the three schemes can be flexibly set by those skilled in the art, and this application does not impose specific limitations on this. For example, the final result can be obtained by weighting the assessment results of the three schemes, or, for example, if any one of the three assessments indicates a risk, then the test chamber is deemed to have a health risk and an early warning is issued.

[0180] As attached Figure 3 As shown, the integrated circuit aging bench high-temperature test chamber heating wire health assessment method of this application embodiment includes the following steps:

[0181] Step 301: Generate sample data based on the historical health operation data of the test chamber under multiple working conditions.

[0182] The integrated circuit aging bench high temperature test chamber heating wire health assessment method provided in this embodiment of the invention is applied to an electronic device. The electronic device includes a processor and a memory. The memory stores an integrated circuit aging bench high temperature test chamber heating wire health assessment program or instruction. When the integrated circuit aging bench high temperature test chamber heating wire health assessment program or instruction is executed by the processor, the integrated circuit aging bench high temperature test chamber heating wire health assessment method flow is realized.

[0183] In one optional embodiment, generating sample data based on the historical health operation data of the test chamber under multiple operating conditions may include the following sub-steps:

[0184] Sub-step 1: Collect historical health operation data of the test chamber under multiple operating conditions.

[0185] The historical health operation data includes: temperature, voltage, current, operating parameters, and indicators to show the temperature control stage of the test chamber. The temperature control stage includes: heating stage, steady-state stage, and cooling stage. Operating parameters may include, but are not limited to: set temperature and ambient temperature.

[0186] In practical implementation, historical health operation data, covering various operating conditions, can be collected during the test chamber's health period. For each operation record, at discrete sampling times:

[0187] t_k, k = 0,1,...,K

[0188] The collected historical health operation data for multi-condition testing includes: temperature (T), voltage (V), current (I), and operating condition information, set temperature (T_set), set humidity (H_set), ambient temperature (T_amb), and program step identifier (step_id). The program step identifier represents the temperature control stage set by the program, including the heating process, steady-state process, and cooling process, denoted as up, flat, and down respectively. This information can be directly obtained, thus allowing for the division of the data into temperature control stages.

[0189] Sub-step 2: Generate derived historical health operation data of the test chamber based on historical health operation data and sampling intervals.

[0190] The derived data includes power and resistance.

[0191] Let the sampling interval be dt, which can be considered a constant. Then the formulas for calculating the power and resistance of the derived data are as follows:

[0192]

[0193] Sub-step 3: Normalize the historical health operation data and derived data to obtain normalized basic data of physical quantities.

[0194] In practical implementation, maximum-minimum normalization can be used for different physical quantities. The normalization formula is as follows:

[0195]

[0196] in, The basic data for normalized physical quantities The data consists of unnormalized physical quantities, i = 1, 2, 3... n.

[0197] In this step, voltage, current, power, resistance, and operating parameters are also normalized and scaled to the [0,1] interval to eliminate the influence of dimensions, thereby improving the stability of network training.

[0198] Sub-step 4: Calculate the average temperature rise slope based on the data from all temperature rise stages under the same operating conditions.

[0199] Sub-step 5: Based on the average temperature rise slope, remove early degraded data from the normalized physical quantity baseline data to generate sample data.

[0200] Sub-steps 4 and 5 involve the sample data screening process. Only by ensuring that the sample data does not contain early degradation fault information can the subsequent comparisons be considered reasonable. Therefore, in a batch of confirmed healthy operating records, the average temperature rise slope is calculated for all temperature rise segments under the same operating conditions. Suspected early degradation sample data is removed using the temperature rise slope. If the difference value is higher than 10%, the data is considered to contain early degradation data and cannot be used for training the health baseline curve. The same operating conditions can include the same target temperature, ambient temperature, etc.

[0201] Step 302: Perform progress normalization processing on the sample data for different temperature control stages and uniform length resampling processing for the same temperature control stage to construct a training sample set.

[0202] The dominant physical mechanisms of test chambers under different operating conditions may differ. Therefore, different constraint weights should be applied to different stages of heating, steady-state cooling, and temperature rise. However, the time series length of each sample is different, and the time series lengths of different stages also vary. Therefore, it is necessary to standardize the sample length. This mainly includes two core steps: normalization of progress at different temperature control stages and resampling with a unified sample length at the same temperature control stage.

[0203] In an optional embodiment, the method of constructing a training sample set by performing progress normalization processing on sample data for different temperature control stages and uniform length resampling processing for the same temperature control stage may include the following sub-steps:

[0204] Sub-step 1: For each temperature control stage, normalize the progress of the temperature control stage.

[0205] The temperature control phase includes: the heating phase, the steady-state phase, and the cooling phase.

[0206] For any temperature control stage interval Define the progress of this temperature control stage. :

[0207]

[0208] in, For the current moment, This is the start time of the temperature control phase. This is the end time of the temperature control phase.

[0209] Sub-step 2: Preset the sampling length for each temperature control stage.

[0210] To facilitate the training of the multi-physical quantity reference curve generation model, a uniform sampling length is set for each temperature control stage in this embodiment:

[0211] This indicates the length of the heating stage; for example, it can be set to 600, meaning 600 points will be sampled during this heating stage. This indicates the length of the steady-state phase and can be set to 1000. This indicates the length of the cooling phase, which can be set to 600.

[0212] It should be noted that the above is only an example illustrating the sampling length of a temperature control stage. In the actual implementation process, those skilled in the art can flexibly set the sampling length values ​​according to actual needs. This application embodiment does not impose specific restrictions on this.

[0213] Sub-step 3: For each temperature control stage, resample on the progress axis according to the sampling length corresponding to the temperature control stage and the normalized temperature control stage progress to obtain the resampled sample data.

[0214] Sub-step 4: Perform linear interpolation on the resampled sample data to obtain a temperature control stage sequence of uniform length corresponding to each physical quantity.

[0215] Sub-step 5: Construct a training sample set based on the temperature control stage sequence of uniform length corresponding to each physical quantity in each temperature control stage.

[0216] Taking the heating phase as an example, the process of resampling on the progress axis can be as follows:

[0217] right Define the progress of the heating phase. :for

[0218]

[0219] Linear interpolation can yield a uniform-length sequence of heating stages, such as a sequence of physical quantity data including temperature, voltage, and current.

[0220]

[0221] The resampling process for the steady-state and cooling phases is similar, ultimately yielding a uniform sequence of temperature control phases for constructing a training sample set.

[0222] Step 303: Train the pre-established multi-physical quantity benchmark curve generation model using the training sample set to obtain the target multi-physical quantity benchmark curve generation model.

[0223] Among them, the multi-physical quantity benchmark curve generation model integrates operating condition feature encoding and Transformer encoder, and can predict the multi-physical quantity health benchmark curves corresponding to different operating conditions. The Transformer encoder is a deep learning model based on the self-attention mechanism.

[0224] In one optional embodiment, training a pre-established multi-physical quantity benchmark curve generation model using a training sample set to obtain the target multi-physical quantity benchmark curve generation model may include the following sub-steps:

[0225] Sub-step 1: Classify the training sample set according to the temperature control stage to obtain the input feature vector sequence corresponding to each temperature control stage.

[0226] Sub-step 2: For each temperature control stage, input the corresponding input feature vector sequence into the working condition feature encoder.

[0227] The working condition characteristic encoder includes two fully connected networks.

[0228] For a certain temperature control stage, such as the heating stage, at the t-th time step, the corresponding input characteristic vector is:

[0229]

[0230] The input characteristic vector sequence corresponding to the entire heating stage is: .

[0231] Sub-step 3: Project the input feature vectors onto the high-dimensional latent feature space through the first fully connected network to obtain the first feature vector sequence. Then, map the first feature vector sequence to the standard hidden dimension through the second fully connected network to obtain the initial hidden sequence.

[0232] Since the input feature vector mainly consists of low-dimensional physical condition parameters and time progress, its data distribution is relatively sparse and lacks high-frequency texture features found in image signals. To fully exploit the complex nonlinear coupling relationships between different condition parameters, this embodiment constructs a condition feature encoder consisting of a two-layer fully connected network. In the specific implementation process:

[0233] First, the input feature vector is mapped through the first layer of linear mapping. Projecting onto a higher-dimensional latent feature space and extracting interaction features using a non-linear activation function:

[0234]

[0235] in, and The learnable weight matrix and biases for the first layer. These are features of the intermediate hidden layer. The resulting sequence is the first eigenvector sequence.

[0236] Secondly, the feature dimensions are mapped to the standard hidden dimensions. :

[0237]

[0238] The final encoded initial hidden sequence is obtained as follows: .

[0239] This encoder can more directly establish a mapping relationship from operating condition commands to high-dimensional state embeddings, providing input representations with rich semantic information for the subsequent global temporal reasoning of the Transformer encoder.

[0240] Sub-step 4: Add positional encoding to the convolutional output of the initial hidden sequence to obtain the second feature vector sequence.

[0241] To incorporate temporal location information, positional encoding is added to the initial hidden sequence:

[0242]

[0243] in, For the first The position encoding vector for each time step can be a fixed trigonometric function encoding or a learnable parameter.

[0244] Sub-step 5: Input the second feature vector sequence into the two-layer Transformer encoder to obtain the target feature vector output by the Transformer encoder.

[0245] Each layer of the Transformer encoder includes: a multi-head self-attention and feedforward network structure;

[0246] Based on this, two Transformer encoder layers are stacked, each layer including a multi-head self-attention and feedforward network structure. The layer's input and output are:

[0247]

[0248] The final output of the Transformer encoder is:

[0249]

[0250] Where, h t ∈R , for time step High-level temporal characteristics representation, h t The resulting sequence is the second feature vector sequence.

[0251] Sub-step 6: Input the target feature vector into the three fully connected headers to obtain the voltage reference curve, temperature reference curve, and current reference curve.

[0252] This sub-step uses three fully connected connectors to output temperature, voltage, and current health reference curves, respectively, as shown below:

[0253]

[0254] Sub-step 7: Based on the voltage reference curve, temperature reference curve and current reference curve, derive the power reference curve and resistance reference curve to form a multi-physical quantity reference curve.

[0255] The following formula can be used to derive the power and resistance health reference curves based on the output temperature, voltage, and current health reference curves:

[0256]

[0257] It should be noted that when calculating the derived power and resistance health reference curves, the output temperature, voltage, and current health reference curves need to be inversely normalized.

[0258] Sub-step 8: Based on the model training mechanism of multi-physical quantity reference curves and temperature control stage physical constraints, train the pre-established multi-physical quantity reference curve generation model to obtain the target multi-physical quantity reference curve generation model.

[0259] In this embodiment of the application, a model training method with physical constraints in different temperature control stages is adopted, which mainly includes two parts: prediction error loss ME and physical constraints weighted by different temperature control stages.

[0260] A feasible method for training a pre-established multi-physical quantity reference curve generation model based on a model training mechanism that utilizes multi-physical quantity reference curves and physical constraints of temperature control stages to obtain a target multi-physical quantity reference curve generation model may include the following sub-steps:

[0261] Sub-step 1: For each temperature control stage, predict the error loss of the real health label corresponding to the temperature control stage in the training sample set.

[0262] For the real health labels corresponding to the temperature control stage in the training sample set, the prediction error loss MSE is defined as follows:

[0263]

[0264]

[0265]

[0266] The error loss of the actual health label corresponding to the temperature control stage is:

[0267] Sub-step 2: Determine the weighted physical constraints for the temperature control stage.

[0268] The physical constraints include: power consistency constraints, resistance consistency constraints, and power-temperature direction consistency constraints.

[0269] In one feasible implementation, the power consistency constraint term can be calculated as follows:

[0270] The power health baseline curve is calculated based on the voltage health baseline curve and current health baseline curve managed in the multi-physical quantity baseline curve generation model; the power baseline curve is calculated based on the voltage baseline curve and current baseline curve obtained by predicting the training samples using the multi-physical quantity baseline curve generation model; the deviation between the power health baseline curve and the power baseline curve is used as a power consistency constraint.

[0271] Power consistency constraint: This is based on the healthy power reference P calculated using the healthy voltage reference curve and the healthy current reference curve. The deviation between them is used as a constraint term to suppress the generation of power curves inconsistent with the voltage and current relationship by the model; specifically, it can be expressed as follows:

[0272]

[0273] In one feasible implementation, the resistance consistency constraint term can be calculated as follows:

[0274] Based on the voltage health reference curve and current health reference curve managed in the multi-physical quantity reference curve generation model, the resistance health reference curve is calculated; based on the voltage reference curve and current reference curve obtained by predicting the training samples using the multi-physical quantity reference curve generation model, the resistance reference curve is calculated; the deviation between the resistance health reference curve and the resistance reference curve is used as a resistance consistency constraint.

[0275] Resistance consistency constraint: The healthy resistance benchmark calculated based on the healthy voltage reference curve and the healthy current reference curve, and the predicted resistance benchmark curve. The deviation between them is used as a constraint to highlight the physical significance of resistance as a sensitive indicator of the degradation of heating wire materials. The specific representation of the resistance consistency constraint can be as follows:

[0276]

[0277] Power and temperature directional consistency constraint: According to physics, there is directional consistency between power and temperature; that is, as power increases, temperature increases. Therefore, physical constraints can be added to the training process to increase interpretability.

[0278] Calculate the discrete changes in reference temperature and power (t≥1):

[0279]

[0280]

[0281] Add physical constraints using directional consistency:

[0282] when and Penalties will be imposed (considering response latency) in the following form:

[0283]

[0284] To reflect the differences in physical mechanisms at different stages, three physical constraint weights are introduced: heating, steady state, and cooling.

[0285]

[0286] For example Take 0.7, Take 0.5, The value is set to 0.3. It should be noted that the specific values ​​of the three physical constraint weights can be flexibly set by those skilled in the art, and no specific restrictions are imposed on them in this embodiment.

[0287] For each time step, based on its temperature control stage label Select the corresponding weights:

[0288]

[0289] The weighted power-temperature direction consistency constraint term during the temperature control stage is expressed as follows:

[0290]

[0291] Sub-step 3: Construct the total loss function of the model based on physical constraints.

[0292] Therefore, in order to simultaneously optimize prediction accuracy and physical consistency, the final total loss function of the physical constraint model is composed of the prediction error loss in sub-step 1 and the three physical constraint losses in sub-step 2. The specific formula for calculating the total loss function is as follows:

[0293]

[0294] Sub-step 4: Based on the total loss function of the model corresponding to each temperature control stage and the training sample set, train the multi-physical quantity benchmark curve generation model to obtain the target multi-physical quantity benchmark curve generation model.

[0295] Step 304: Input the operating parameters of the test chamber to be evaluated into the target multi-physical quantity reference curve generation model to predict the health reference curves of each physical quantity.

[0296] Steps 301 to 303 describe the training process of the multi-physical quantity benchmark curve generation model. After the training model is completed and the target multi-physical quantity benchmark curve generation model is generated, the trained model can be applied to the system platform where the test chamber to be evaluated is located. The system platform evaluates the health status of the heating wire of the test chamber to be evaluated based on the model. Steps 304 to 308 describe the specific process of health assessment and health index calculation of the heating wire in the test chamber.

[0297] The physical quantities include: temperature, voltage, current, resistance, and power.

[0298] This step generates multi-physical quantity health baseline curves online. When the test chamber to be evaluated is actually running, the current operating conditions are input into the trained target multi-physical quantity baseline curve generation model to obtain the curves of each physical quantity, which are then used as the multi-physical quantity health baseline curves of the heating wire under the current operating conditions.

[0299]

[0300] Step 305: Calculate the actual power and resistance using the real-time temperature, real-time voltage, and real-time current of the test chamber to be evaluated.

[0301] The real-time temperature, voltage, and current of the test chamber under evaluation, collected in real time, are shown below:

[0302]

[0303] The actual power and resistance calculated based on real-time temperature, voltage, and current are expressed as follows:

[0304]

[0305] Step 306: Within the specified actual window, use the fast dynamic time warping algorithm to calculate the target physical quantity channel sequence distance based on real-time temperature, actual power, resistance, and the target physical quantity health baseline curve.

[0306] The target physical quantities include temperature, power, and resistance. This step assesses the sequence similarity between the measured physical quantity health curve and the health baseline curve.

[0307] In practical implementation, the distance between multiple channels, i.e., the distance between the target physical quantity channels, can be calculated using a fast dynamic time warping algorithm within a specified time window.

[0308]

[0309] Step 307: Calculate the weight of each target physical quantity based on the information entropy corresponding to each target physical quantity.

[0310] Steps 307-308 are the calculation process of the health index using the entropy weight method. Step 307 first calculates the weight corresponding to each target physical quantity, and step 308 performs a weighted sum based on the weight corresponding to each target physical quantity and the channel sequence distance to determine the health index of the heating wire of the test chamber to be evaluated.

[0311] In this step, the health index is obtained by weighted fusion of the DTW output values ​​of the three target physical quantities. Weights are calculated using a large amount of historical data collected offline and applied to the online phase. The formula is as follows:

[0312] Suppose we have j indicators, ,in Let n be the number of samples. Then, perform max-min normalization to obtain the matrix. .

[0313]

[0314]

[0315] Calculate the information entropy of each indicator:

[0316]

[0317] Calculate the weight of each indicator:

[0318]

[0319] Where k is the number of indicators.

[0320] Step 308: Determine the health index of the heating wire in the test chamber to be evaluated based on the channel sequence distance and weight corresponding to each target physical quantity.

[0321] In one optional embodiment, determining the health index of the heating wire in the test chamber to be evaluated based on the channel sequence distance and weight corresponding to each target physical quantity may include the following sub-steps:

[0322] Sub-step 1: Based on the channel sequence distance and weight corresponding to each target physical quantity, calculate the degradation degree of the heating wire of the test chamber to be evaluated;

[0323] An exemplary formula for calculating the overall degradation degree is as follows:

[0324]

[0325] in, The DTW distance for the j-th target physical quantity is the channel sequence distance.

[0326] Sub-step 2: Use an exponential function to map the degree of degradation to a health index within a preset range.

[0327] The preset range value can be flexibly set by those skilled in the art, and there are no specific restrictions on this in the embodiments of this application. For example, it can be set to 0-100.

[0328] The degree of degradation can be mapped to a health index HI of 0-100 using an exponential function based on the following formula:

[0329]

[0330] in, HI is an adjustment coefficient used to adjust the sensitivity of the health score according to actual working conditions. When D is 0, HI of 100 indicates that the current health status of the test chamber resistance wire is completely consistent with the baseline; as the difference increases, HI gradually decreases. The complete consistency of the current health status of the test chamber resistance wire with the baseline can also be understood as the complete coincidence of the currently acquired physical quantity curves characterizing the health status of the test chamber resistance wire with the health baseline curves of each target physical quantity determined by the model and matched to the current working conditions.

[0331] The integrated circuit aging bench high-temperature test chamber heating wire health assessment method provided in this invention has the following aspects: First, it introduces multi-physical quantity information to construct a multi-channel health benchmark curve under dynamic operating conditions, so that the degradation characteristics of the heating wire are reflected not only in the temperature response, but also in the changes in electrical power and resistance, which can improve the sensitivity to gradual degradation behavior. Second, before training the multi-physical quantity benchmark curve generation model, training samples with an average temperature rise slope significantly lower than the healthy level are removed, which can effectively avoid early degradation data from contaminating the health benchmark curve, thereby improving the reliability of the benchmark. Third, it introduces staged weighted power-temperature physical constraints in the model training, so that the physical constraint strength of different temperature control stages can be set separately, which conforms to the differences in the heating dominant mechanism of different stages, and can improve the physical rationality and interpretability of the health benchmark curve. Fourth, it combines multi-channel DTW sequence similarity and adaptively determines the weight of each target physical quantity through the entropy weight method to output the final health index, and the assessment results are intuitive and reliable.

[0332] As attached Figure 4 As shown, the health assessment method for the fan of the high-temperature test chamber of the integrated circuit aging bench according to an embodiment of this application includes the following steps:

[0333] Step 401: Collect monitoring parameters of the test chamber to be evaluated.

[0334] The monitoring parameters include: temperature, vibration frequency, and fan speed detected by multiple sensors inside the test chamber.

[0335] The integrated circuit aging bench high temperature test chamber fan health assessment method provided in this embodiment of the invention is applied to an electronic device. The electronic device includes a processor and a memory. The memory stores an integrated circuit aging bench high temperature test chamber fan health assessment program or instructions. When the integrated circuit aging bench high temperature test chamber fan health assessment program or instructions are executed by the processor, the integrated circuit aging bench high temperature test chamber fan health assessment method flow is realized.

[0336] In one optional embodiment, the method of acquiring the monitoring signal of the test chamber to be evaluated may include the following sub-steps:

[0337] Sub-step 1: Pre-set N temperature sensors inside the test chamber to be evaluated.

[0338] Among them, N temperature sensors cover the front, middle and rear areas of the test chamber to be evaluated in the horizontal direction, and cover the upper, middle and lower three layers of the test chamber to be evaluated in the vertical direction.

[0339] In practical implementation, temperature sensors can be arranged inside the high-temperature test chamber based on the nine-point temperature measurement method, with N set to 9. The temperature sensors are arranged inside the test chamber according to the nine-point temperature measurement method, forming a three-dimensional nine-square grid layout, simultaneously collecting nine-point temperature measurement signals, vibration frequency signals, and fan speed signals. Temperature data can be collected every 5 seconds, with a temperature sampling accuracy of ±0.1℃.

[0340] Sub-step 2: Test the temperature measurement accuracy of N temperature sensors under different test chamber temperatures and load conditions. If the test results meet the set conditions, the test is considered passed.

[0341] This step is a test procedure for the stability and reliability of temperature measurement using N sensors. Once the test is passed, the signal acquisition and monitoring process will begin. It should be noted that sub-steps 1 and 2 are not executed every time a monitoring signal is acquired; they only need to be executed once when N temperature sensors are deployed.

[0342] Sub-step 3: Collect the temperatures of the N preset temperature sensors in the test chamber to be evaluated to obtain N temperature rise curves, and collect the vibration frequency and fan speed of the test chamber to be evaluated.

[0343] Preferably, the temperature signals collected from the N sensors can be used to determine whether they meet the preset uniformity requirement by calculating the difference between the maximum and minimum temperature values ​​of the N temperature sensors under stable conditions. If the requirement is met, the collected temperature signals are considered valid. The preset uniformity can be flexibly set by those skilled in the art, and this embodiment does not impose specific limitations on it; for example, the uniformity can be set to ≤2℃.

[0344] In practical implementation, an example of sensor deployment and signal acquisition process can be as follows:

[0345] The temperature uniformity within the test chamber was evaluated using a nine-point temperature measurement method. Nine temperature probes were placed at specific locations within the test chamber to collect real-time data at different temperature points, as detailed below:

[0346] (1) Principles of probe placement

[0347] Based on the spatial distribution inside the test chamber, nine temperature probes are arranged in a "three-dimensional nine-square grid" position inside the test chamber: horizontally covering the front, middle, and rear areas inside the chamber; vertically covering the upper, middle, and lower layers; typical positions include: one temperature sensor in the center of the test chamber, one temperature sensor at the midpoint of each side (4 in total), and one temperature sensor at each of the four corners (4 in total), ensuring coverage of the main space inside the test chamber and avoiding the omission of temperature deviations caused by dead zones in fan circulation.

[0348] (2) Testing process

[0349] Temperature point selection: Four typical high-temperature points of 100℃, 125℃, 150℃, and 175℃ were selected to cover the commonly used operating range of the equipment. Anlan AL1000 temperature recorder was used to collect temperature data of each probe every 5 seconds, and each temperature point was continuously maintained for 2 hours to accumulate a large amount of time-series data. Tests were conducted under two load conditions: "empty box" and "filled with 16 aging boards" to simulate the impact of samples on air circulation in actual use. For example, when the aging boards were filled, the temperatures of probes 4 and 5 were lower, which confirmed the interference of the load on temperature uniformity.

[0350] (3) Evaluation indicators of acquired signals

[0351] Temperature signals from the temperature sensors are collected, and the difference between the maximum and minimum temperature values ​​of the nine probes under steady-state conditions is calculated to determine whether the requirement of "uniformity ≤ 2℃" is met. If it is met, the nine temperature sensors are deemed to be stable and suitable for temperature detection within the test chamber.

[0352] Step 402: Extract features from the monitoring parameters to generate a high-dimensional temperature rise curve feature vector.

[0353] In one optional embodiment, the method for extracting features from the monitoring signal to generate a high-dimensional temperature rise curve feature vector can be as follows:

[0354] For each of the N temperature rise curves, extract the preset features of the temperature rise curve, integrate the preset features corresponding to the N temperature rise curves, and generate a high-dimensional temperature rise curve feature vector.

[0355] The preset features may include, but are not limited to: the linear fitting slope during the temperature rise phase, the peak value during the temperature steady-state phase, the time required for the temperature to reach the preset temperature, the temperature rise area, and the temperature fluctuation.

[0356] In the specific implementation process, the temperature rise curve of each temperature signal includes the entire process from room temperature to the set temperature. The following features are extracted from the temperature rise curve of each temperature signal.

[0357] The linear fitting slope (k) during the temperature rise phase: Temperature rise phase (0-t) i The slope of the linear fit, inversely

[0358] The heating rate is expressed by the formula:

[0359] Among them, t i T is the sampling time. i For the corresponding temperature, These are the mean values ​​of time and temperature, respectively, where n is...

[0360] Number of sampling points during the temperature rise phase;

[0361] Peak value during the steady-state temperature phase (T) max ): The average temperature during the stable phase (temperature fluctuation ≤ ±0.5℃);

[0362] Arrival time (t) 80 %) Temperature reaches 0.8T max The moment;

[0363] Temperature rise area (S): The integral area enclosed by the temperature rise curve and the time axis, calculated using the following formula: Where t2 is the start time of the stable phase;

[0364] Temperature fluctuation (σ): The standard deviation of temperature during the steady-state phase, expressed by the formula: , among which, T j For the j-th temperature value in the steady-state phase, The average temperature during the steady-state phase is given by m, where m is the number of sampling points during the steady-state phase.

[0365] Temperature uniformity (ΔT): The maximum temperature difference at the same time, for example, 9 temperature sensors (N channels), is given by the formula: .

[0366] Step 403: Input the fan speed, chamber environment parameters, and vibration frequency of the test chamber to be evaluated into the pre-built digital mirror observer to obtain the virtual mirror temperature field of the test chamber to be evaluated.

[0367] In this step, a digital mirror observer is pre-built. The application of digital mirror observer technology based on digital twins in high-temperature test chambers differs from other fields. It does not directly monitor the equipment itself, but rather constructs a high-fidelity "component-environment" coupled physical model. Using the macroscopic state of the environment as a sensor, it inversely infers and senses early, invisible performance degradation of core components. Its ultimate goal is to ensure the absolute reliability and safety of the internal processes. The digital mirror observer pre-built in this step is a fan-enclosure coupled reduced-order digital mirror model, with the core objective of indirectly assessing the health status of internal components through the macroscopic state of the environment.

[0368] The constructed digital mirror observer includes: a mechanistic model and LSTM (Long Short-Term Memory Networks); the method of obtaining the virtual mirror temperature field of the test chamber by inputting the fan speed, chamber environmental parameters, and vibration frequency of the test chamber to be evaluated into the pre-constructed digital mirror observer can include the following sub-steps:

[0369] Sub-step 1: Input the fan speed, chamber environment parameters, and vibration frequency of the test chamber to be evaluated into the pre-built digital mirror observer;

[0370] Sub-step 2: The mechanism model is based on fan speed and chamber environment parameters to obtain the theoretical temperature field of the test chamber to be evaluated;

[0371] In this step, the physical parameters of the fan during the healthy phase, such as the rated speed n, are collected. o Fan blade diameter d, motor resistance R o ; Enclosure parameters such as thermal resistance R t Heat capacity C t The temperature field distribution data is used as a prototype database to construct a mechanism model.

[0372] The inputs to the mechanistic model are fan speed and enclosure environmental parameters. The mechanistic model is based on the heat conduction equation.

[0373] A temperature field simulation model was built to obtain the theoretical temperature field T. mech The heat conduction equation is:

[0374]

[0375] Where ρ is the air density, c is the specific heat capacity of air, λ is the thermal conductivity, and Q is the air density. f For fan cooling, Q f Q is directly proportional to the fan speed n. f = k a n, k aThe proportionality constant is used. The environmental parameters of the enclosure include: ρ (air density), c (specific heat capacity of air), and λ (thermal conductivity). The above differential equation is solved using the finite difference method, outputting the theoretical temperature field vector T with nine nodal temperature values. mech .

[0376] Sub-step 3: LSTM determines the temperature field deviation compensation value based on fan speed and vibration frequency;

[0377] This sub-step is a data-driven fusion operation, specifically constructing and offline training an LSTM residual compensation model. The input is fan speed and vibration frequency. Through training, the LSTM learns to fit nonlinear errors that the underlying mechanism model cannot describe. In online applications, it outputs the temperature field deviation compensation value ΔT. error It is used to fit the small changes in thermal efficiency caused by nonlinear errors and wear.

[0378] Sub-step 4: Weighted summation of the theoretical temperature field and the temperature field deviation compensation value to obtain the virtual mirror temperature field of the test chamber to be evaluated.

[0379] Theoretical temperature field T mech The output of the mechanistic model is the temperature field deviation compensation value ΔT. error The output of the LSTM is used; the LSTM can be considered a data model. The virtual mirror temperature field of the test chamber to be evaluated can be obtained by weighted fusion of the outputs of the two models using the following formula:

[0380]

[0381] The weight corresponding to the mechanism model is K1, and the weight corresponding to the data model is K2. The specific values ​​of K1 and K2 can be flexibly set by those skilled in the art. In this embodiment, no specific restrictions are imposed. For example, K1=0.7 and K2=0.3 can be set.

[0382] Using this mirror image as an "ideal reference state," the resulting virtual mirror temperature field T virtual It provides state prediction values ​​to support subsequent data fusion work, assists in verifying the validity of sensor data, helps generate health data under multiple operating conditions during the baseline construction stage, and provides residual analysis basis for subsequent health calculation.

[0383] In this optional embodiment, a real-time health assessment of the temperature chamber fan of the high-temperature aging test chamber is performed based on a digital mirror observer. From the minute changes in the temperature field, the early degradation of fan performance can be keenly captured. In this specific equipment, the high-temperature aging test chamber, a "non-invasive, indirect" health assessment of key auxiliary components can help testers to grasp the current status of the equipment in a timely manner, discover potential hidden dangers in advance, and ensure that subsequent aging tests are carried out as planned.

[0384] Step 404: Determine the feature vector of the real high-dimensional temperature rise curve based on the feature vector of the high-dimensional temperature rise curve and the virtual mirror temperature field.

[0385] In practical implementation, the virtual mirror temperature field and the feature vector of the sensor's measured temperature rise curve can be weighted and fused to obtain a denoised high-fidelity temperature rise curve. Subsequently, features such as slope and peak value are extracted from this curve to obtain the feature vector X of the true high-dimensional temperature rise curve. real In this scenario, the feature vector was formed by extracting six features from each temperature rise curve of the nine temperature sensors, and then stitching together all the features to create X. real (9, 6).

[0386] Step 405: Using a multi-distance metric fusion algorithm, the health status of the test chamber fan is calculated based on the health status data generated from the real high-dimensional temperature rise curve feature vector, the health status data generated from the historical full life cycle feature sequence of the sample test chamber, the mean vector of health stage features, and the health stage feature covariance matrix.

[0387] In an optional embodiment, a multi-distance metric fusion algorithm is used to calculate the health status of the test chamber fan based on the real high-dimensional temperature rise curve feature vector, the health status data generated from the historical full life cycle feature sequence of the sample test chamber, the health stage feature mean vector, and the health stage feature covariance matrix. The method may include the following sub-steps:

[0388] Sub-step 1: Calculate the Mahalanobis distance based on the feature vector of the real high-dimensional temperature rise curve, the mean vector of the health stage features, and the covariance matrix of the health stage features.

[0389] The Mahalanobis distance is used to indicate the deviation between the real-time characteristics of the fan in the test chamber being evaluated and the healthy baseline.

[0390] Mahalanobis distance (d): reflects the deviation of real-time features from the healthy baseline distribution, and is expressed by the formula:

[0391]

[0392] Among them, X real Σ is the feature vector of the true high-dimensional temperature rise curve, μ is the feature mean vector of the healthy stage, and Σ is the feature covariance matrix of the healthy stage.

[0393] In practical implementation, the historical full lifecycle feature sequence of the sample test chamber can be screened using TCN. The screened health stage data is used as the input source, and statistical analysis is performed on all feature vectors of this stage to obtain μ. The mean vector μ is calculated by taking the arithmetic mean of all vectors in the same dimension of the health sample set, representing the feature center of the equipment in its optimal state. Based on μ, the correlation and fluctuation range between features are calculated to obtain Σ, representing the feature fluctuation range and correlation of the equipment in its optimal state.

[0394] Sub-step 2: Based on the feature vector of the real high-dimensional temperature rise curve and the mean vector of the health stage features, calculate the cosine similarity between the real-time features of the fan of the test chamber to be evaluated and the health baseline.

[0395] Cosine similarity (Sim_c): reflects the consistency of feature vector directions, and can be calculated using the following formula:

[0396]

[0397] Sub-step 3: Based on the feature vector of the real high-dimensional temperature rise curve and the mean vector of the health stage features, calculate the Manhattan distance between the real-time features of the fan of the test chamber to be evaluated and the health baseline.

[0398] Manhattan distance (d m ): Reflects local feature mutations, and the calculation formula can be:

[0399]

[0400] In the formula, d is m The Manhattan distance value, where D is the total dimension of the feature vectors (e.g., 54). For real-time feature vector X real The i-th element in Let be the i-th element in the health mean vector μ.

[0401] Sub-step 4: Use a multi-distance metric fusion algorithm to fuse Mahalanobis distance, cosine similarity, and Manhattan distance to obtain the health status H of the test chamber fan.

[0402] In this embodiment, the health status H of the test chamber fan is calculated based on a multi-distance metric fusion algorithm.

[0403] The health fusion formula can be (H∈[0,1]):

[0404]

[0405] In the formula, The weighting coefficients for each distance metric (must be greater than zero, and can be set based on experience). The value is cosine distance, because the closer the cosine similarity is to 1, the healthier it is. Here it is converted into "distance / penalty term".

[0406] During the equipment health phase, data is collected for various typical operating condition combinations (e.g., 100°C empty chamber, 150°C full load, etc.). An independent health baseline is constructed for each operating condition, forming a baseline library. The health baseline is a statistical abstraction of all characteristic data during the health phase and serves as a benchmark for measuring the current state of the equipment.

[0407] In this embodiment, the interpretability analysis of the health status of the test chamber fan can also be performed. Specifically, the SHAP method can be used to calculate the contribution of each feature to the decline in health status, and the output format is as follows:

[0408]

[0409] Where C i Let H be the contribution of the i-th feature, H be the health score, and x be the weight of the i-th feature. i This represents the value of the i-th feature acquired in real time. The contribution of each feature can be calculated using the above formula. In practical implementation, the contribution weights of the first N features to the decline in health H can be output to help pinpoint the main cause of degradation. The specific value of N can be flexibly set by those skilled in the art; this application does not impose specific limitations on this, for example, N can be 3, 2, or 4.

[0410] In an optional embodiment, the mean vector of health stage features and the covariance matrix of health stage features can be constructed in the following manner, specifically including the following sub-steps:

[0411] S1: Obtain the historical full life cycle feature sequence of the sample test chamber.

[0412] S2: Add labels to a portion of the data in the historical full lifecycle feature sequence to generate a training sample set.

[0413] The label is used to represent the corresponding health status as healthy or deteriorated. A label of 0 can represent healthy, and a label of 1 can represent deteriorated.

[0414] S3: Optimize the temporal convolutional network by performing depthwise separable convolution and model pruning to obtain a lightweight temporal convolutional network.

[0415] S4: Train the lightweight temporal convolutional network using the training sample set to generate the target temporal convolutional network.

[0416] This application embodiment uses a lightweight temporal convolutional network (TCN) to identify degradation points in test chamber fans. The lightweight TCN is obtained by optimizing the TCN with depthwise separable convolution and pruning. The lightweight TCN is used to identify degradation points in the historical full-lifecycle feature sequences of the sample test chambers. First, the PELT change point detection algorithm is used for preliminary analysis of the feature sequences to locate "suspected points" where data distribution abruptly changes, which are then the focus areas for the TCN. A training sample set is input into the lightweight TCN, and the model learns the feature boundaries of healthy and degraded states, supporting deployment on edge computing nodes.

[0417] In practical implementation, the TCN training process also includes a testing phase. A portion of the training sample set is used for training, while the remaining samples are used to test the prediction accuracy of the trained TCN. During testing, the feature sequences corresponding to the training samples are input into the TCN. The TCN then determines the corresponding degradation initiation point. The determination criteria for the degradation initiation point are optimized as follows: the TCN output "degradation probability" exceeds 0.4 for three consecutive time windows, and its moving average first exceeds 0.5; this moment is marked as the "degradation initiation point." The TCN output results are compared with the health status marked on the training samples to determine the prediction accuracy of the trained TCN. Once the prediction accuracy meets the preset conditions, the TCN training is considered complete, and the target TCN is obtained.

[0418] S5: Use a target temporal convolutional network to perform time series analysis on all data in the historical full life cycle feature sequence to determine the degradation starting point when the sample test chamber changes from a healthy state to a degraded state.

[0419] S6: Based on the degradation initiation point, determine the health status data and degradation status data in the historical full life cycle feature sequence.

[0420] The target TCN performs time-series analysis on all data in the historical full lifecycle feature sequence, and the output results accurately divide the historical data into two parts:

[0421] Health phase data, or health status data (before the onset of degradation), is used to generate a health baseline.

[0422] Degradation phase data, i.e. degradation state data (after the degradation initiation point): is used to verify the sensitivity of TCN.

[0423] The core function of TCN degradation point identification is to accurately mark the critical moment when a device transitions from a "healthy" state to a "degraded" state. It clearly divides the device's entire lifecycle data into different degradation stages. This provides a clean data source for building subsequent health baselines, preventing interference during baseline construction. The lightweight TCN design, while maintaining temporal feature extraction capabilities, is more suitable for real-time operation on resource-constrained edge devices through techniques such as model compression and knowledge distillation.

[0424] S7: Generate the mean vector of health stage features and the covariance matrix of health stage features based on health status data.

[0425] Step 406: Issue a fan health status warning based on health level.

[0426] In one optional embodiment, the method for providing fan health status early warning based on health level can be as follows:

[0427] When the health status of the fan in the test chamber to be evaluated is lower than the warning threshold and higher than the shutdown threshold, an early warning message will be output.

[0428] When the health status of the fan in the test chamber to be evaluated reaches the shutdown threshold, the fan will be shut down and an early warning will be issued.

[0429] When the health status of the fan in the test chamber to be evaluated is higher than the safety threshold, update the mean vector of the health stage features and the covariance matrix of the health stage features.

[0430] The warning message may include the contribution of health status and degradation characteristics. The safety threshold is greater than the warning threshold, and the warning threshold is greater than the shutdown threshold. The specific values ​​of these three thresholds can be flexibly set by those skilled in the art, and this embodiment does not impose specific limitations on them. For example, the warning threshold H_warn can be set to 0.7, corresponding to a 15% performance degradation; the shutdown threshold H_stop can be set to 0.3, corresponding to a 40% performance degradation; and the safety threshold can be set to 0.9.

[0431] A feasible way to update the mean vector and covariance matrix of health stage features is as follows:

[0432] When the health status of the fan in the test chamber to be evaluated is higher than the safety threshold, the eigenvector of the true high-dimensional temperature rise curve and the mean vector of the health stage feature are weighted and summed to obtain the updated mean vector of the health stage feature. The covariance matrix of the health stage feature is updated based on the eigenvector of the true high-dimensional temperature rise curve and the updated mean vector of the health stage feature to obtain the updated covariance matrix of the health stage feature.

[0433] Baseline Update: To eliminate baseline drift caused by seasonal changes in ambient temperature and to enable baseline adaptability through exponentially weighted moving average, an update is performed daily when the device is deemed healthy (H > safety threshold, e.g., 0.9). The specific formula used for the update can be as follows:

[0434]

[0435]

[0436] The integrated circuit aging bench high-temperature test chamber fan health assessment method provided in this application embodiment collects monitoring parameters of the test chamber to be assessed; extracts features from the monitoring parameters to generate a high-dimensional temperature rise curve feature vector; inputs the fan speed, chamber environment parameters, and vibration frequency of the test chamber to be assessed into a pre-constructed digital mirror observer to obtain a virtual mirror temperature field of the test chamber to be assessed; determines the real high-dimensional temperature rise curve feature vector based on the high-dimensional temperature rise curve feature vector and the virtual mirror temperature field; uses a multi-distance metric fusion algorithm to calculate the health level of the test chamber fan based on the real high-dimensional temperature rise curve feature vector, the health status data generated from the historical full life cycle feature sequence of the sample test chamber, and the health stage feature mean vector and health stage feature covariance matrix; and provides a fan health status early warning based on the health level. The method provided in this application embodiment, based on digital twin digital mirror technology, constructs a real-time dynamic mapping between physical entities and virtual models to achieve data interaction and closed-loop optimization between physical and virtual spaces. For the incubator fan of the test chamber, the digital mirror observer uses the physical entities of the fan and the test chamber as prototypes, and integrates the mechanism model, operating characteristics, real-time sensor data and historical operation and maintenance data to build a virtual copy. It can accurately simulate the fan's operating status and predict the fan's health degradation trend, thereby realizing online quantification and early warning of the health status of the incubator fan in the high temperature test chamber.

[0437] Figure 5 The structural block diagram of the integrated circuit high-temperature aging bench sealing performance degradation state assessment device according to the embodiments of this application is shown.

[0438] The integrated circuit high-temperature aging test bench sealing performance degradation assessment device provided in this application includes the following functional modules:

[0439] The first generation module 501 is used to collect temperature information from various temperature sensors arranged in a preset area around the door panel of the high-temperature aging test chamber and generate a temperature change curve over time.

[0440] The feature extraction module 502 is used to extract features from the temperature change curves of each temperature sensor over time to obtain the feature sequence of the heating curves of each temperature sensor.

[0441] The first estimation module 503 is used to input the characteristic sequence of each heating curve and the historical temperature sequence of each temperature sensor into a pre-established data-driven model to obtain the first temperature distribution vector and leakage parameter estimate for each location. The first temperature distribution vector contains the temperature prediction value for each location at the next moment. Each temperature sensor corresponds to one location.

[0442] The prediction module 504 is used to input each of the temperature feature sequences into a pre-established physical driving model to predict the second temperature distribution vector at each of the locations;

[0443] The second generation module 505 is used to generate a temperature distance consistency evaluation index based on the second temperature distribution vector and the actual measured temperature distribution vectors of each of the collected locations.

[0444] The target state vector determination module 506 is used to obtain the target state vector based on the first temperature distribution vector, the second temperature distribution vector and the temperature distance consistency evaluation index;

[0445] The second estimation module 507 is used to recursively estimate the target state vector to obtain the leakage coefficient estimate at each of the locations.

[0446] Calculation module 508 is used to calculate the sealing health index of the high-temperature aging bench test chamber based on the estimated leakage coefficient at each of the locations;

[0447] The determination module 509 is used to determine the sealing performance of the high-temperature aging test chamber door panel based on the sealing health index and the estimated leakage coefficient at each of the locations.

[0448] Optionally, the feature extraction module is specifically used for:

[0449] For each temperature sensor, the initial heating rate, equivalent time constant, area of ​​temperature curve, and spatial temperature gradient features are extracted to generate a heating curve feature sequence corresponding to the temperature sensor.

[0450] Optionally, the first estimation module includes:

[0451] The first submodule is used to generate a first time-series feature vector corresponding to each of the positions based on the feature sequences of each of the heating curves and the historical temperature sequences of each of the temperature sensors.

[0452] The second submodule is used to input each of the first temporal feature vectors into the TCN module of the data-driven model for convolution operation to obtain the second temporal feature vectors corresponding to each position.

[0453] The third submodule is used to perform nonlinear mapping on each of the second time-series feature vectors input into the KAN module of the data-driven model to obtain the temperature prediction value of each location at the next time step.

[0454] The fourth submodule is used to determine the estimated value of the leakage parameter at each location based on the predicted temperature value at the next moment and the actual measured temperature value at the location.

[0455] Optionally, the second generation module is specifically used for:

[0456] Calculate the gradient consistency index, shape similarity index, and energy ratio consistency index of the first temperature distribution vector and the actual measured temperature distribution vectors collected at each of the aforementioned locations;

[0457] The gradient consistency index, shape similarity index, and energy ratio consistency index are weighted and summed to generate a temperature distance consistency evaluation index.

[0458] Optionally, the target state vector determination module is specifically used for:

[0459] Calculate the observation noise based on the temperature-distance consistency evaluation index;

[0460] The target state vector is obtained by summing the first temperature distribution vector, the second temperature distribution vector, and the observation noise.

[0461] Optionally, the determination module is specifically used for:

[0462] If the sealing health index is less than a first preset threshold, it is determined that the overall sealing performance of the high-temperature aging test chamber door panel has deteriorated.

[0463] For each of the locations, the estimated leakage coefficient at that location is compared with a second preset threshold to filter out target locations with local sealing abnormalities.

[0464] Optionally, the device further includes:

[0465] The analysis module is used to perform time series trend analysis on the temperature distance consistency evaluation index to obtain trend change results;

[0466] The early warning module is used to determine that the door panel of the high-temperature aging test chamber has early deterioration when the trend change result indicates a continuous downward trend, and output an early warning prompt.

[0467] In the embodiments of this application Figure 5The integrated circuit high-temperature aging bench sealing performance degradation assessment device shown can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit it.

[0468] The embodiments provided in this application Figure 5 The integrated circuit high-temperature aging test bench sealing performance degradation assessment device shown can achieve Figure 1 The various processes implemented in the method implementation examples will not be described again here to avoid repetition.

[0469] Optionally, embodiments of this application also provide an electronic device, including a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the processes executed by the aforementioned integrated circuit high-temperature aging bench sealing performance degradation state evaluation device and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0470] It should be noted that the electronic device in this application embodiment includes the server described above.

[0471] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0472] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0473] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for evaluating the degradation state of sealing performance on a high-temperature aging test bench for integrated circuits, characterized in that, The method includes: The temperature information of each temperature sensor deployed in a preset area around the door panel of the high-temperature aging test chamber is collected, and a temperature change curve over time is generated. Feature extraction is performed on the temperature change curves of each temperature sensor over time to obtain the feature sequence of the temperature rise curves corresponding to each temperature sensor. The characteristic sequences of each heating curve and the historical temperature sequences of each temperature sensor are input into a pre-established data-driven model to obtain the first temperature distribution vector and leakage parameter estimate for each location. The first temperature distribution vector contains the predicted temperature value for each location at the next moment. Each temperature sensor corresponds to one location. The characteristic sequences of each of the heating curves are input into a pre-established physical driving model to predict the second temperature distribution vector at each of the locations. Based on the second temperature distribution vector and the actual measured temperature distribution vectors collected at each of the aforementioned locations, a temperature distance consistency evaluation index is generated; Based on the first temperature distribution vector, the second temperature distribution vector, and the temperature distance consistency evaluation index, the target state vector is obtained; The target state vector is recursively estimated to obtain the leakage coefficient estimate at each of the locations; The sealing health index of the high-temperature aging test chamber is calculated based on the estimated leakage coefficient at each of the aforementioned locations. The sealing performance of the high-temperature aging test chamber door is determined based on the sealing health index and the estimated leakage coefficient at each of the aforementioned locations.

2. The method according to claim 1, characterized in that, The step of extracting features from the temperature-time change curves corresponding to each of the temperature sensors to obtain the feature sequence of the temperature rise curves corresponding to each of the temperature sensors includes: For each temperature sensor, the initial heating rate, equivalent time constant, area of ​​temperature curve, and spatial temperature gradient features are extracted to generate a heating curve feature sequence corresponding to the temperature sensor.

3. The method according to claim 1, characterized in that, The steps of inputting the characteristic sequences of each of the heating curves and the historical temperature sequences of each of the temperature sensors into a pre-established data-driven model to obtain the predicted temperature value and the estimated leakage parameter value at each location at the next moment include: Based on the characteristic sequence of each heating curve and the historical temperature sequence of each temperature sensor, a first temporal feature vector corresponding to each position is generated; Each of the first temporal feature vectors is input into the TCN module of the data-driven model for convolution operation to obtain the second temporal feature vector corresponding to each position; Each of the second time-series feature vectors is input into the KAN module of the data-driven model for nonlinear mapping to obtain the temperature prediction value at the next time step for each of the locations; For each of the locations, the estimated leakage parameters at that location are determined based on the predicted temperature value at the next moment and the actual measured temperature value at that location.

4. The method according to claim 1, characterized in that, The step of generating a temperature distance consistency evaluation index based on the second temperature distribution vector and the actual measured temperature distribution vectors collected at each of the aforementioned locations includes: Calculate the gradient consistency index, shape similarity index, and energy ratio consistency index of the second temperature distribution vector and the actual measured temperature distribution vectors collected at each of the aforementioned locations; The gradient consistency index, shape similarity index, and energy ratio consistency index are weighted and summed to generate a temperature distance consistency evaluation index.

5. The method according to claim 1, characterized in that, The step of obtaining the target state vector based on the first temperature distribution vector, the second temperature distribution vector, and the temperature distance consistency evaluation index includes: Calculate the observation noise based on the temperature-distance consistency evaluation index; The target state vector is obtained by summing the first temperature distribution vector, the second temperature distribution vector, and the observation noise.

6. The method according to claim 1, characterized in that, The step of determining the sealing performance of the aging bench temperature test chamber door panel based on the sealing health index and the estimated leakage coefficient at each of the aforementioned locations includes: If the sealing health index is less than a first preset threshold, it is determined that the overall sealing performance of the high-temperature aging test chamber door panel has deteriorated. For each of the locations, the estimated leakage coefficient at that location is compared with a second preset threshold to filter out target locations with local sealing abnormalities.

7. The method according to claim 1, characterized in that, The method further includes: Time series trend analysis was performed on the temperature distance consistency evaluation index to obtain the trend change results; When the trend change result indicates a continuous downward trend, it is determined that the door panel of the high-temperature aging test chamber has early deterioration and an early warning is output.

8. A device for evaluating the sealing performance degradation status of an integrated circuit high-temperature aging test bench, characterized in that, The device includes: The first generation module is used to collect temperature information from various temperature sensors arranged in a preset area around the door panel of the high-temperature aging test chamber and generate a temperature change curve over time. The feature extraction module is used to extract features from the temperature change curves of each temperature sensor over time to obtain the feature sequence of the heating curves of each temperature sensor. The first estimation module is used to input the characteristic sequence of each of the heating curves and the historical temperature sequence of each of the temperature sensors into a pre-established data-driven model to obtain the first temperature distribution vector and the leakage parameter estimate for each location. The first temperature distribution vector contains the temperature prediction value for each location at the next moment. Each temperature sensor corresponds to one location. The prediction module is used to input the feature sequences of each of the heating curves into a pre-established physical driving model to predict the second temperature distribution vector at each of the locations; The second generation module is used to generate a temperature distance consistency evaluation index based on the second temperature distribution vector and the actual measured temperature distribution vectors collected at each of the locations. The target state vector determination module is used to obtain the target state vector based on the first temperature distribution vector, the second temperature distribution vector, and the temperature distance consistency evaluation index. The second estimation module is used to recursively estimate the target state vector to obtain the leakage coefficient estimate at each of the locations. The calculation module is used to calculate the sealing health index of the high-temperature aging bench test chamber based on the estimated leakage coefficient at each of the aforementioned locations. The determination module is used to determine the sealing performance of the high-temperature aging test chamber door panel based on the sealing health index and the estimated leakage coefficient at each of the locations.

9. The apparatus according to claim 8, characterized in that, The feature extraction module is specifically used for: For each temperature sensor, the initial heating rate, equivalent time constant, area of ​​temperature curve, and spatial temperature gradient features are extracted to generate a heating curve feature sequence corresponding to the temperature sensor.

10. The apparatus according to claim 8, characterized in that, The first estimation module includes: The first submodule is used to generate a first time-series feature vector corresponding to each of the positions based on the feature sequences of each of the heating curves and the historical temperature sequences of each of the temperature sensors. The second submodule is used to input each of the first temporal feature vectors into the TCN module of the data-driven model for convolution operation to obtain the second temporal feature vectors corresponding to each position. The third submodule is used to perform nonlinear mapping on each of the second time-series feature vectors input into the KAN module of the data-driven model to obtain the temperature prediction value of each location at the next time step. The fourth submodule is used to determine the estimated value of the leakage parameter at each location based on the predicted temperature value at the next moment and the actual measured temperature value at the location.

Images