A method for rapidly evaluating online refining effect of an automobile die-casting aluminum alloy melt

Abstract

The application discloses a kind of online refining effect fast evaluation methods of automobile die-casting aluminum alloy melt, specifically related to melt refining quality monitoring field, the method includes: through acoustic sensor, thermal imager and electrochemical sensor, synchronous acquisition refining process multimodal sensing data;Parallel processing various data streams, extract acoustic features characteristic of melt viscous property, temperature field characteristics of thermal uniformity and electromotive force characteristics of alumina concentration change;Using time series fusion network based on attention mechanism, analyze the dynamic coupling relationship of multiple parameters, and adaptively calculate the refining effect index;Compare the dynamic threshold interval generated by historical data clustering, realize real-time evaluation and feedback control of refining effect, and output the decision instruction of adjusting refining gas flow, stirring speed or holding power.The application realizes multidimensional, high-precision online monitoring and rapid evaluation, significantly improves the quality control efficiency and stability of the melt refining process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of melt refining effect evaluation, and more particularly to a method for rapidly evaluating online refining effect of automobile die-casting aluminum alloy melt. BACKGROUND

[0002] In the production process of automobile die-casting aluminum alloy, melt refining is a key link to ensure the quality of the final casting, and its purpose is to remove gas and inclusions in the melt and adjust the chemical composition to improve the mechanical properties and casting properties of the alloy. At present, the evaluation of refining effect in the industry mainly relies on offline testing and manual experience judgment. This method has obvious hysteresis and cannot guide the production process in real time. Moreover, the sampling process may introduce secondary pollution. In recent years, some online monitoring technologies have been gradually applied to the aluminum alloy melting process, such as using a single sensor to monitor a specific parameter. However, in actual use, there are still some shortcomings, such as: on the one hand, single-parameter monitoring cannot fully reflect the multi-physical field coupling characteristics of the refining process; secondly, the existing methods lack a multi-source information fusion mechanism, and different sensor data are not synchronized in time and not associated in physical meaning, resulting in one-sided evaluation results and insufficient reliability.

[0003] On the other hand, most systems only provide monitoring functions without forming a closed-loop control, and cannot adjust the refining parameters in real time according to the evaluation results, still relying on manual intervention. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a method for rapidly evaluating online refining effect of automobile die-casting aluminum alloy melt, which solves the problems in the above background art by the following scheme.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a method for rapidly evaluating online refining effect of automobile die-casting aluminum alloy melt, comprising: S1: synchronous acquisition of multi-modal sensor data: synchronously and in real time acquiring multi-modal sensor data of the melt in the refining process collected by an acoustic sensor, a thermal imager and an electrochemical sensor;

[0006] The multi-modal sensor data includes acoustic sensor data, thermal imaging data and electrochemical sensor data;

[0007] S2: parallel calculation of physical state parameters: parallel processing of the multi-modal sensor data, obtaining acoustic characteristic parameters based on the acoustic sensor data, calculating temperature field characteristic parameters based on the thermal imaging data, and calculating electromotive force characteristic parameters based on the electrochemical sensor data;

[0008] S3: Refining effect fusion evaluation: analyze the dynamic coupling relationship between the acoustic characteristic parameter, the temperature field characteristic parameter and the electromotive force characteristic parameter and its mode of change over time, and calculate the refining effect index representing the refining quality and stability of the melt;

[0009] S4: Dynamic decision and feedback: real-time monitor the change trend of the refining effect index, and compare it with the dynamic threshold interval generated based on the historical qualified melt data statistics, output the evaluation result of the refining effect and the corresponding control instruction.

[0010] Preferably, the acoustic sensing data is a broadband acoustic emission signal measured by an acoustic sensor below the melt liquid surface, reflecting the escape and rupture characteristics of refining gas bubbles;

[0011] The thermal imaging data is a two-dimensional infrared thermal image sequence measured by a thermal imager in the melt surface area, reflecting the temperature spatial distribution characteristics;

[0012] The electrochemical sensing data is an electromotive force signal measured by an electrochemical sensor inserted into the melt, reflecting the change of alumina concentration in the melt.

[0013] Preferably, the acoustic characteristic parameter is analyzed based on the acoustic sensing data, specifically including:

[0014] The time-frequency domain transformation is performed on the broadband acoustic emission signal, and the ratio of the signal energy in the preset characteristic frequency band to the total energy is calculated, and the ratio is taken as the acoustic characteristic parameter representing the viscosity characteristics of the melt.

[0015] Preferably, the temperature field characteristic parameter is calculated based on the thermal imaging data, specifically including:

[0016] For each frame of image in the two-dimensional infrared thermal image sequence, the maximum temperature gradient value and the temperature standard deviation of the melt surface region of interest are calculated, and the weighted sum of the maximum temperature gradient value and the temperature standard deviation is taken as the temperature field characteristic parameter representing the thermal uniformity of the melt.

[0017] Preferably, the electromotive force characteristic parameter is calculated based on the electrochemical sensing data, specifically including:

[0018] The electromotive force signal is filtered and denoised, and the fluctuation frequency and average fluctuation amplitude of the electromotive force signal per unit time are calculated, and the product of the fluctuation frequency and the average fluctuation amplitude is taken as the electromotive force characteristic parameter representing the change of alumina concentration in the melt.

[0019] Preferably, the dynamic coupling relationship analysis process includes:

[0020] A nonlinear mapping model between the acoustic characteristic parameter, the temperature field characteristic parameter and the electromotive force characteristic parameter is established, the time delay cross-correlation function between each two parameters is calculated, the interaction intensity and the causal time sequence between the bubble behavior, the heat transfer and the electrochemical reaction are quantified, and the interaction intensity is taken as the core input for calculating the refining effect index.

[0021] Preferably, the calculation of the refining effect index is based on a dynamic evaluation model of a time sequence fusion network based on an attention mechanism.

[0022] The model dynamically allocates the contribution of the acoustic, thermal imaging and electrochemical three characteristic parameters to the final index calculation at each moment through attention weight, so as to adaptively capture the dominant influencing factors at different stages of the refining process.

[0023] Preferably, the generation mode of the dynamic threshold interval is to perform cluster analysis on the multi-batch refining effect index curves of the historical qualified melt refining process, extract the index change envelope representing the excellent refining process, and define the upper limit and lower limit of the envelope as the dynamic threshold interval.

[0024] Preferably, the output refining effect evaluation result and the corresponding control instruction specifically include:

[0025] When the refining effect index deviates from the dynamic threshold interval, a decision signal with parameter adjustment priority is generated according to which one of the three characteristic parameters at the current moment first deviates abnormally and the deviation direction;

[0026] The decision signal includes a suggestion to adjust the refining gas flow, the stirring speed or the melt holding power.

[0027] The technical effects and advantages of the present application are:

[0028] 1. The present application realizes multi-dimensional real-time monitoring of the melt viscosity characteristics, thermal uniformity and change of alumina concentration through synchronous acquisition and parallel processing of acoustic, thermal imaging and electrochemical multi-modal sensing data, effectively overcomes the defects of one-sidedness and insufficient reliability of the existing single parameter monitoring technology, and significantly improves the comprehensiveness and accuracy of the refining process state perception.

[0029] 2. The present application quantifies the dynamic coupling relationship between the bubble behavior, the heat transfer and the electrochemical reaction by establishing a multi-parameter nonlinear mapping model and time delay cross-correlation analysis, and adaptively calculates the refining effect index by using a fusion network based on an attention mechanism, which can accurately capture the changes of the dominant factors at different refining stages and realize fine evaluation of the refining quality and stability.

[0030] 3. This invention generates dynamic threshold intervals based on historical data clustering and designs a parameter adjustment priority decision mechanism, thereby realizing real-time evaluation and closed-loop feedback control of refining effect. This effectively reduces reliance on human experience, improves the automation and adaptability of refining process control, and ultimately significantly improves the quality control efficiency and product consistency of automotive die-cast aluminum alloy melt. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the overall structure of the present invention;

[0032] Figure 2 This is a schematic diagram of the multimodal sensing data synchronous acquisition process of the present invention;

[0033] Figure 3 This is a schematic diagram of the parallel solution process for the physical state parameters of the present invention;

[0034] Figure 4 This is a schematic diagram of the dynamic decision-making and feedback process of the present invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] As attached Figures 1-4 The method shown is a rapid evaluation method for online refining effect of automotive die-cast aluminum alloy melt, including:

[0037] S1: Synchronous acquisition of multimodal sensing data: synchronously and in real time acquires multimodal sensing data of the melt during the refining process collected by acoustic sensors, thermal imagers and electrochemical sensors;

[0038] The multimodal sensing data includes acoustic sensing data, thermal imaging data, and electrochemical sensing data;

[0039] It should be specifically noted that the acoustic sensing data is a broadband acoustic emission signal measured by an acoustic sensor below the molten liquid surface, reflecting the characteristics of the escape and collapse of refining gas bubbles.

[0040] The thermal imaging data is a sequence of two-dimensional infrared thermal images of the surface region of the melt, which are measured by a thermal imager and reflect the spatial distribution characteristics of temperature.

[0041] The electrochemical sensing data is the electromotive force signal of the change in alumina concentration in the reaction melt, measured by an electrochemical sensor inserted into the melt.

[0042] It needs to be further explained that the acoustic sensor selects PCB356A15 type high temperature resistant acoustic emission sensor, the range is 10-1000 kHz, the highest tolerance temperature reaches 260℃; It is fixed on the side wall of the smelting furnace through the ceramic insulating seat, the sensor probe is stretched into the smelting liquid below 5-8 cm, so as to avoid the interference of the furnace wall vibration, and meanwhile, the sensor probe can ensure to capture the bubble sound, the sensor probe is wrapped with 0.5 mm thick alumina ceramic sleeve, the sleeve is not in direct contact with the smelting, the gap is ≤1 mm, the ceramic acoustic impedance is 38*10 6 kg / (m²・s), and the matching coefficient of the acoustic impedance with the ADC12 aluminum alloy smelting is ≥0.8, the signal is connected to the data acquisition module through the shielding cable with the length ≤5 m; The acoustic signal acquisition adopts NI9234 module, 4 channels, 24 bit resolution, the highest sampling rate is 100 kHz, the sampling frequency is 50 kHz, and the sampling range is ±10 V.

[0043] The thermal imager selects FLIRA655sc type infrared thermal imager, the resolution is 640*512 pixels, the temperature measurement range is 600-750℃, covering the smelting temperature 650-700℃ of the ADC12 aluminum alloy, and the frame frequency is 30 fps; It is fixed on the bracket 1.8-2.2 m above the furnace body, the lens axis is perpendicular to the smelting liquid surface, the outside is additionally provided with a quartz dust cover, the smelting surface is preset as a region of interest (ROI) through software, and only the smelting surface data is collected in each frame of image; The thermal imaging data acquisition adopts NI9221 module with 16 bit resolution, the temperature data of each frame of image is acquired in real time through the analog output interface of the thermal imager, and the time stamp is generated synchronously when each frame of image is collected.

[0044] The electrochemical sensor adopts Al-Si-Cu alloy electrode consistent with the smelting composition, the diameter is 8 mm, the length is 150 mm, the outside is sleeved with high temperature resistant quartz tube, only 5 mm probe is exposed, and the non-target reaction between the electrode and the smelting is reduced; It is inserted into the smelting 6-10 cm through the reserved hole of the furnace cover, the electrode lead is connected to the CHI660E electrochemical workstation through the high temperature resistant lead wire above 300℃, and the sampling frequency of the workstation is set to 10 Hz; The electrochemical signal acquisition communicates with the electrochemical workstation through the NI9842 serial module, and the electromotive force signal is synchronously connected to the collection system at the frequency of 10 Hz.

[0045] It needs to be further explained that the synchronization checking mechanism triggers the “time stamp calibration” every 15 seconds, compares the time stamp deviation of the three types of data, if the deviation exceeds 2 ms, the clock of the collection module is automatically adjusted, and the data is aligned in the time dimension.

[0046] It needs to be further explained that the acoustic signal adopts 50Hz notch filter + db4 wavelet base 5 layer threshold denoising, the threshold is calculated through Birgé-Massart strategy, and the bubble sound characteristics are reserved; The thermal imaging data adopts 3*3 median filter to remove salt and pepper noise + histogram equalization to enhance temperature contrast; The electrochemical signal adopts 4-order Butterworth low-pass filter with a cutoff frequency of 1 Hz to remove high-frequency electromagnetic noise + linear trend removal by least square fitting baseline to eliminate the influence of electrode temperature drift.

[0047] S2: Parallel solution of physical state parameters: parallel processing of the multi-modal sensing data, analyzing acoustic characteristic parameters based on acoustic sensing data, calculating temperature field characteristic parameters based on thermal imaging data, and calculating electromotive force characteristic parameters based on electrochemical sensing data.

[0048] It needs to be specifically explained that the acoustic characteristic parameters are analyzed based on the acoustic sensing data, which specifically includes:

[0049] The wideband acoustic emission signal is transformed in time and frequency domain, and the ratio of signal energy in the preset characteristic frequency band to total energy is calculated, which is taken as the acoustic characteristic parameter representing the melt viscosity characteristics.

[0050] It needs to be further explained that in the acoustic signal processing, first, the short-time Fourier transform is performed on the denoised signal, Hanning window is adopted, window length is 1024 points, corresponding to 0.02048 seconds, and 50% overlap rate is set, to obtain the power spectrum density matrix , where f is the frequency range 20Hz-20kHz, t is the time, , which represents the acoustic signal power at frequency f at time t, and is used to reflect the frequency distribution of acoustic energy; then, according to the calibration of the previous experiment, 1000Hz to 5000Hz is selected as the characteristic frequency band, the energy of this frequency band is mainly derived from bubble collapse, which is closely related to the melt viscosity, and the full frequency band range is set as 20Hz-20000Hz; the characteristic frequency band energy and the total energy are calculated respectively according to the formula and , the integral adopts trapezoidal method, step , and GPU parallel acceleration is used to ensure the calculation efficiency; finally, the acoustic characteristic parameter is defined as the ratio of the two , the value range is [0, 1], the higher the value, the stronger the bubble collapse sound in 1-5kHz frequency band, the lower the melt viscosity, and vice versa; through 50 batches of ADC12 melt test calibration, the parameter range is about [0.3, 0.6] under normal refining conditions.

[0051] It needs to be specifically explained that the temperature field characteristic parameters are calculated based on the thermal imaging data, which specifically includes:

[0052] For each frame image in the sequence of two-dimensional infrared thermal images, the maximum temperature gradient value and the temperature standard deviation of the melt surface region of interest are calculated, and a weighted sum of the maximum temperature gradient value and the temperature standard deviation is taken as a temperature field characteristic parameter representing the thermal uniformity of the melt.

[0053] It should be further explained that the melt surface region of interest is located at the center of the melt surface, and the pixel size is 400*400, corresponding to an actual physical size of 50cm*50cm, calibrated with a focal length of 15mm and an installation height of 2m, and 1 pixel is about 1.25mm, avoiding the interference of the edge temperature caused by the heat dissipation of the furnace wall; then the maximum temperature gradient wherein T(i,j,t) represents the temperature of the (i,j)th pixel in the ROI at time t, which is directly measured by the thermal imager, d represents the physical distance between adjacent pixels, both being 1.25mm, The smaller the value is, the more gentle the temperature change in the ROI is, and the better the thermal uniformity is; the temperature standard deviation reflects the dispersion degree of the temperature distribution in the ROI, and its calculation formula is wherein N is the total number of pixels, Tavg represents the current average temperature, The smaller the value is, the more concentrated the temperature is; finally, a temperature field characteristic parameter is obtained by weighting the maximum temperature gradient and the temperature standard deviation wherein w1 and w2 are weight coefficients, and 5 groups of combinations are selected, compared with the offline thermal uniformity detection results, and finally , .

[0054] It should be further explained that, because is of the order of magnitude of , 1m is multiplied by to convert it into ℃ / m, so that the overall unit of the result is ℃ / m; after calibration, the value of under normal refining conditions is in the range of [50,150]℃ / m.

[0055] It should be specifically explained that the electromotive force characteristic parameter calculated based on the electrochemical sensing data specifically comprises:

[0056] The electromotive force signal is filtered and denoised, the fluctuation frequency and the average fluctuation amplitude of the electromotive force signal per unit time are calculated, and the product of the fluctuation frequency and the average fluctuation amplitude is taken as an electromotive force characteristic parameter representing the change of the concentration of aluminum oxide in the melt.

[0057] ​​​It should be further explained that in the electromotive force signal processing, the raw signal acquired by the electrochemical workstation is first processed. Preprocessing was performed using a 4th-order Butterworth low-pass filter with a cutoff frequency of 1Hz to remove high-frequency noise, resulting in... Then, the linear trend term is removed by least squares fitting to eliminate the baseline shift caused by electrode temperature drift, finally obtaining a drift-free signal. Then, two key characteristic quantities were calculated: one is the fluctuation frequency. Defined as the electromotive force signal exceeding a threshold per second. The number of effective peak values; the threshold setting was calibrated through 30 batches of experiments. When the change in alumina concentration caused an electromotive force fluctuation exceeding 0.05V, it was considered an effective fluctuation. The value is set to 0.05; peak detection is performed by detecting local maxima within a 1-second sliding window containing 10 sampling points and comparing them with the window mean. The difference reflects the frequency of concentration changes; secondly, the average fluctuation amplitude. Its calculation formula is ,in The number of valid peak values ​​within the current window, and equal, For the k-th peak voltage, A larger value indicates a higher fluctuation intensity and a more significant change in alumina concentration; finally, the characteristic parameters of the electromotive force are calculated through the fluctuation frequency and the average fluctuation amplitude. Its value range is approximately [0.02, 0.8] V*Hz under normal refining conditions. The larger the value, the higher the alumina concentration in the melt. Since the change in alumina content will affect the conductivity of the melt, it will lead to an increase in the fluctuation of electromotive force.

[0058] S3: Refining effect fusion evaluation: Analyze the dynamic coupling relationship between acoustic characteristic parameters, temperature field characteristic parameters and electromotive force characteristic parameters and their time-varying patterns, and calculate the refining effect index that characterizes the refining quality and stability of the melt.

[0059] It should be specifically noted that the dynamic coupling relationship analysis process includes:

[0060] A nonlinear mapping model is established among acoustic characteristic parameters, temperature field characteristic parameters, and electromotive force characteristic parameters. By calculating the time-delay cross-correlation function between each pair of parameters, the interaction strength and causal sequence among bubble behavior, heat transfer, and electrochemical reaction are quantified, and this interaction strength is used as the core input for calculating the refining effect index.

[0061] It should be further explained that the nonlinear mapping model is constructed based on bubble behavior. Heat transfer Electrochemical reactions For input, a ternary nonlinear mapping model is constructed

[0062]

[0063] where Y is the coupling relationship representation value, is a nonlinear function fitted by a three-layer BP neural network, with 3 neurons in the input layer, 10 neurons in the hidden layer, and 1 neuron in the output layer, and the activation function is ReLU, is an error term; this network is trained using 500 sets of experimental data, ; then the dynamic coupling relationship between different parameters is quantified, and the time delay cross-correlation function between each pair of parameters is calculated:

[0064]

[0065] where is the time delay, and N is the length of the time series, , and , are the mean and standard deviation of the two parameters, respectively; The absolute value reflects the coupling strength, and positive and negative indicate the causal time sequence; finally, the maximum cross-correlation coefficient between each pair of parameters , and are extracted as the core features representing the interaction strength, which are used for weight distribution in the subsequent fusion network.

[0066] It should be specifically noted that the calculation of the refining effect index is through a dynamic evaluation model composed of a time series fusion network based on an attention mechanism;

[0067] This model dynamically allocates the contribution of acoustic, thermal imaging, and electrochemical three feature parameters to the final index calculation at each time point through attention weights, thereby adaptively capturing the dominant influencing factors at different stages of the refining process.

[0068] It should be further noted that the network structure is divided into 4 layers based on the PyTorch framework, with the input layer dimension being [100, 6], where 100 is the length of the time series and 6 is the input feature , , and three coupling strength features , , The convolution layer uses 2 convolution kernels with a size of 3*1, a quantity of 16, and a ReLU activation function, to extract local features in the time series, with an output dimension of [98, 16]; the attention layer splices the convolution layer output and the coupling strength feature to generate an attention score through a fully connected layer )、 、 with an input of 19 dimensions and an output of 3 dimensions, and then a weight coefficient is obtained through softmax normalization , 、 Similarly, each weight satisfies ; the final fusion layer calculates the refining effect index ; the network is trained using 5000 groups of multi-source parameter-offline quality score data, the batch size of network training is 32, the iteration stop condition is that the loss value is stable and lower than 0.01 for 5 consecutive iterations, and the weights are initialized using He normal; the offline score is based on the inclusion content and density analyzed by metallography, the inclusion content is graded according to GB / T15749-2020, 1 level corresponds to 60 points, 0 level corresponds to 100 points, and the weight of the score is 60%; the density is detected according to GB / T20975.3-2008, the standard density corresponds to 40 points, and each decrease of 0.01 g / cm³ deducts 5 points, accounting for 40% of the score weight; the loss function uses mean square error:

[0069]

[0070] where M is the number of samples, is the offline score; the Adam optimizer is used for training, the learning rate is set to 0.001, and the iteration is performed for 1000 times until the loss is lower than 0.01; finally, is linearly mapped to the range of 0-100, and the higher the value represents the better the refining effect, and the value of in the normal refining process is in the range of [60, 90].

[0071] S4: Dynamic decision and feedback: real-time monitoring of the change trend of the refining effect index, and comparison with the dynamic threshold interval generated based on the historical qualified melt data statistics, output of the evaluation result of the refining effect and the corresponding control instruction.

[0072] It needs to be specifically pointed out that the generation method of the dynamic threshold interval is to perform cluster analysis on the multi-batch refining effect index curves of the historical qualified melt refining process, extract the index change envelope line representing the excellent refining process, and define the upper limit and lower limit of the envelope line as the dynamic threshold interval.

[0073] It needs to be further pointed out that to generate the dynamic threshold interval, first collect the The curve data is 600 seconds long for each batch, the sampling interval is 1 second, and the abnormal mutation values caused by equipment failure and other factors are removed by the 3σ principle, and finally 95 batches of effective curves are retained for subsequent analysis; then, based on the K-means clustering method, the dynamic threshold value at each time is extracted, and for each time , the RI value at this time is extracted from the 95 batches of curves to form a data set , set the cluster number K=3, initialize three cluster centers, and then classify by calculating the Euclidean distance from the data points to the center, and iteratively update the cluster center until it is stable; select the category with the most data, extract the maximum value and the minimum value in the category , to form the dynamic threshold interval , ] at this time; in order to adapt to the slow changes of the melt composition and equipment state in actual production, the system also has a threshold updating mechanism, and every time 20 batches of qualified data are added, the above clustering analysis is performed again to update the dynamic threshold interval, so as to maintain the adaptability and robustness of the model.

[0074] It needs to be specifically pointed out that the evaluation results of the output refining effect and the corresponding control instructions specifically include:

[0075] When the refining effect index deviates from the dynamic threshold interval, according to which one of the three characteristic parameters at the current time deviates first and its deviation direction, a decision signal with parameter adjustment priority is generated;

[0076] The decision signal includes suggestions for adjusting the refining gas flow, stirring speed or melt holding power.

[0077] It needs to be further pointed out that the decision signal with parameter adjustment priority is generated, first the abnormal judgment logic is executed: the current is compared with the dynamic threshold interval , in real time; if is within the dynamic threshold interval, output "refining is normal" and do not adjust the parameters; if < or > , the values of the three types of parameters , , in the last 5 seconds are traced back to identify the first "abnormal parameter" that exceeds the normal range; then according to the type of abnormal parameter, in the order of priority, alumina concentration > thermal uniformity > viscosity, a control instruction is generated and sent to the PLC controller through industrial Ethernet:

[0078] If the abnormal parameter is That is, indicates that the concentration of alumina is too high, then increase the holding power, the adjustment amount is Wherein the proportional coefficient 50 is calibrated by experiment;

[0079] If the abnormal parameter is That is, Indicates that the thermal uniformity is poor, then increase the stirring speed, the adjustment amount is ;

[0080] If the abnormal parameter is That is, Indicates that the viscosity is too high, then increase the refining gas flow, the adjustment amount is .

[0081] After the control command is executed, the system continues to monitor Change, if 30 seconds Return to normal threshold interval, output "restore normal"; Otherwise, abnormal tracing is carried out again and the parameters are adjusted until the refining state returns to normal.

[0082] Secondly: the drawings in the disclosed embodiments of the present application only involve the structures involved in the disclosed embodiments, other structures can refer to the usual design, under the condition of no conflict, the same embodiment and different embodiments of the present application can be combined with each other;

[0083] Finally: the above only describes the preferred embodiments of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for rapid evaluation of the online refining effect of automotive die-cast aluminum alloy melt, characterized in that, include: S1: Synchronous acquisition of multimodal sensing data: synchronously and in real time acquires multimodal sensing data of the melt during the refining process collected by acoustic sensors, thermal imagers and electrochemical sensors; The multimodal sensing data includes acoustic sensing data, thermal imaging data, and electrochemical sensing data; S2: Parallel calculation of physical state parameters: The multimodal sensing data is processed in parallel, acoustic characteristic parameters are obtained based on acoustic sensing data analysis, temperature field characteristic parameters are calculated based on thermal imaging data, and electromotive force characteristic parameters are calculated based on electrochemical sensing data. S3: Refining effect fusion evaluation: Analyze the dynamic coupling relationship between acoustic characteristic parameters, temperature field characteristic parameters and electromotive force characteristic parameters and their time-varying patterns, and calculate the refining effect index that characterizes the refining quality and stability of the melt. The dynamic coupling relationship analysis process includes: A nonlinear mapping model is established between acoustic characteristic parameters, temperature field characteristic parameters and electromotive force characteristic parameters. By calculating the time-delay cross-correlation function between pairs of parameters, the interaction strength and causal sequence among bubble behavior, heat transfer and electrochemical reaction are quantified, and the interaction strength is used as the core input for calculating the refining effect index. The calculated refining effect index is evaluated using a dynamic evaluation model based on an attention-based time-series fusion network. The model dynamically allocates the contribution of three feature parameters—acoustic, thermal imaging, and electrochemical—to the final index calculation at each moment by using attention weights, thereby adaptively capturing the dominant influencing factors at different stages of the refining process. S4: Dynamic Decision-Making and Feedback: Monitor the changing trend of the refining effect index in real time, compare it with the dynamic threshold range generated based on historical qualified melt data, and output the evaluation results of the refining effect and the corresponding control instructions. The dynamic threshold interval is generated by performing cluster analysis on the refining effect index curves of multiple batches of historical qualified melt refining processes, extracting the index change envelope that characterizes the excellent refining process, and defining the upper and lower limits of the envelope as the dynamic threshold interval.

2. The method for rapid evaluation of online refining effect of automotive die-cast aluminum alloy melt according to claim 1, characterized in that: The acoustic sensing data is a broadband acoustic emission signal measured by an acoustic sensor below the melt surface, reflecting the characteristics of the escape and collapse of refining gas bubbles. The thermal imaging data is a sequence of two-dimensional infrared thermal images of the surface region of the melt, which are measured by a thermal imager and reflect the spatial distribution characteristics of temperature. The electrochemical sensing data is the electromotive force signal of the change in alumina concentration in the reaction melt, measured by an electrochemical sensor inserted into the melt.

3. The method for rapid evaluation of online refining effect of automotive die-cast aluminum alloy melt according to claim 1, characterized in that: The acoustic feature parameters obtained based on acoustic sensing data analysis specifically include: The broadband acoustic emission signal is transformed in the time and frequency domains, and the ratio of signal energy to total energy within a preset characteristic frequency band is calculated. This ratio is then used as an acoustic characteristic parameter to characterize the viscous properties of the melt.

4. The method for rapid evaluation of online refining effect of automotive die-cast aluminum alloy melt according to claim 1, characterized in that: The temperature field characteristic parameters calculated based on thermal imaging data specifically include: For each frame in the two-dimensional infrared thermogram sequence, the maximum temperature gradient value and temperature standard deviation of the region of interest on the melt surface are calculated, and the weighted sum of the maximum temperature gradient value and temperature standard deviation is used as the temperature field feature parameter characterizing the thermal uniformity of the melt.

5. The method for rapid evaluation of online refining effect of automotive die-cast aluminum alloy melt according to claim 1, characterized in that: The electromotive force characteristic parameters calculated based on electrochemical sensing data specifically include: The electromotive force signal is filtered and denoised, and the fluctuation frequency and average fluctuation amplitude of the electromotive force signal per unit time are calculated. The product of the fluctuation frequency and the average fluctuation amplitude is used as the electromotive force characteristic parameter characterizing the change of alumina concentration in the melt.

6. The method for rapid evaluation of online refining effect of automotive die-cast aluminum alloy melt according to claim 1, characterized in that: The evaluation results of the output refining effect and the corresponding control instructions specifically include: When the refining effect index deviates from the dynamic threshold range, a decision signal with parameter adjustment priority is generated based on which of the three feature parameters deviates abnormally first and its direction of deviation at the current moment. The decision signals include recommendations to adjust the refining gas flow rate, stirring speed, or melt holding power.

Images