A method for quality monitoring in a modified bitumen production process

Abstract

The application provides a quality monitoring method in a modified asphalt production process, and relates to the technical field of quality monitoring.The method comprises the following steps: acquiring multi-point temperature data and stirring torque data in a modified asphalt melting and blending tank, combining real-time angle positions of stirring blades, and constructing a temperature-torque space-time coupling data set; extracting a torque data time sequence and a temperature fluctuation curve with respect to blade phases from the temperature-torque space-time coupling data set, performing fast Fourier transform on the torque data time sequence and the temperature fluctuation curve respectively, and performing frequency domain coupling to obtain a modifier dispersion state characteristic vector; mapping the modifier dispersion state characteristic vector to a phase space coordinate system, identifying system state transition nodes, and constructing a quality state transition prediction function; and based on an output result of the quality state transition prediction function, acquiring an optimal stirring parameter combination, generating an adaptive quality regulation instruction, and realizing an upgrade from passive lag detection to active forward-looking regulation, so that the quality stability and batch consistency of the modified asphalt continuous production are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of quality monitoring technology, and more specifically, to a quality monitoring method in the production process of modified asphalt. Background Technology

[0002] Modified asphalt, as a high-performance road engineering material, is widely used in the construction of high-grade highways, bridges, and special-function pavements. Its excellent high-temperature stability, low-temperature crack resistance, and fatigue resistance make it irreplaceable in the road industry. With the continuous increase in traffic flow and load intensity, the requirements for the quality of modified asphalt and the control of its production process are becoming increasingly stringent. However, the production of modified asphalt is a complex thermophysical and chemical process, and its properties are influenced by many factors, including the type of polymer modifier, dispersion method, matching of production equipment parameters, and control of process conditions. Therefore, the gradual development of online real-time detection and quality control technologies has become an urgent problem to be solved in the field of modified asphalt preparation. In recent years, academia and industry have conducted extensive research on the detection of modifier dispersion, the evaluation and optimization of mixing uniformity during production, and have attempted to monitor the production process using torque sensing, temperature detection, and optical imaging. However, traditional methods have certain limitations in terms of data acquisition dimensions and accuracy, especially in effectively capturing global information on the dispersion state of modifiers under complex dynamic conditions.

[0003] In existing technologies, quality monitoring of the production process mainly relies on the collection and evaluation of single physical quantities, such as single-point temperature fluctuation analysis or simple extraction of torque characteristics. These methods typically cannot comprehensively reflect the dynamic distribution and variation patterns of modifiers in molten base asphalt. Furthermore, while offline detection-based process optimization methods can experimentally verify the dispersion quality of modifiers, their long detection cycles make them unsuitable for the real-time control requirements of modern production. On the other hand, techniques that use single characteristic parameters for feedback adjustment often overlook the significant impact of coupling effects on the uniformity of modifier dispersion, resulting in limited control accuracy. Building on this, existing research has further attempted to combine multidimensional time series analysis techniques with modeling methods, but most have not fully considered the dynamic interaction between key physical quantities such as torque and temperature, as well as their global distribution characteristics. This deficiency not only limits further improvements in production efficiency but also makes it difficult to meet the stringent quality requirements of high-performance asphalt materials. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a quality monitoring method for the production process of modified asphalt. This method can, to some extent, solve the problems of incomplete analysis of the dispersion state of the modifier due to the inability to accurately capture the dynamic coupling characteristics of temperature and torque during the melt blending process of modified asphalt, and the lack of a real-time quality control mechanism based on state transitions, which leads to insufficient precision in optimizing production parameters.

[0005] According to one aspect of the present invention, a quality monitoring method is provided in the production process of modified asphalt, comprising: Multiple temperature and stirring torque data were obtained in the modified asphalt melt blending tank. Combined with the real-time angle position of the stirring blades, a spatiotemporal coupled dataset of temperature and torque was constructed through a spatiotemporal alignment mechanism. The time series of torque data and the fluctuation curve of temperature with blade phase are extracted from the temperature-torque spatiotemporal coupling dataset. After being subjected to fast Fourier transform, they are coupled in the frequency domain to generate a global coupled energy distribution vector. The key feature parameters are extracted and weighted to obtain the feature vector of the modifier dispersion state. The feature vector of the dispersion state of the modifier is mapped to the phase space coordinate system, and the system state transition nodes are identified by the topological changes of the phase space orbits, thereby constructing a mass state transition prediction function. Based on the output of the mass state transition prediction function, the optimal combination of stirring parameters is obtained through parameter optimization and adjustment, and an adaptive quality control command is generated.

[0006] Furthermore, the spatiotemporal alignment mechanism binds the instantaneous temperature values ​​of each temperature acquisition point to the current angular position of the stirrer blade based on the blade phase, constructing a three-dimensional data structure. The first dimension is indexed according to the spatial distribution of temperature sensors, the second dimension is periodically arranged according to the angular position of the stirrer blade, and the third dimension is arranged according to the data acquisition time sequence.

[0007] Furthermore, the Fast Fourier Transform includes a correction process for the original transform result, the correction formula of which is expressed as:

[0008] in, This is the corrected amplitude value. For frequency index, The normalization coefficient for the spectral amplitude is... The magnitude of the original Fast Fourier Transform result. The Gaussian window attenuation intensity parameter, This is the frequency index corresponding to the characteristic frequency of the stirrer. This is the frequency window width parameter. This is the frequency offset suppression coefficient. This is the frequency index corresponding to the basic operating frequency of the mixer.

[0009] Furthermore, the frequency domain coupling is achieved by calculating the conjugate product of the frequency domain characteristic spectrum of the torque data and the frequency domain characteristic spectrum of the temperature-phase fluctuation, as expressed by the formula:

[0010] in, This indicates taking the conjugate complex number. The frequency domain characteristic spectrum of temperature-phase fluctuation at the spatial measurement point. This is the frequency domain characteristic spectrum of torque data.

[0011] Furthermore, the phase space coordinate system is a three-dimensional rectangular coordinate system, and the eigenvector values ​​are used as... Axis coordinates, timestamp as The axis coordinates, the difference between the feature vectors of the current time and the previous time, are used as... The system state transition nodes are determined by combining the detection of local maxima of track turning angles and curvature with track pattern recognition.

[0012] Furthermore, the quality state transition prediction function includes a time-related sub-model, a geometric feature-related sub-model, and a historical state sequence-related sub-model. By linearly combining dynamic weight coefficients and adding a logarithmic time correction term, it outputs the probability value of a quality state transition, expressed by the following formula:

[0013] in, Let be the probability of a mass state transition. For transition sensitivity parameters, To predict the dispersion quality state grade of the modifier, This represents the threshold for quality state transitions. The cumulative distribution function of the standard normal distribution. For orbital geometric characteristics correlation function, and These are the mean and standard deviation parameters of the geometric features, respectively.

[0014] Furthermore, the parameter optimization adjustment determines the first and second sensitive parameters through single-parameter sensitivity analysis, verifies the selection of joint optimization or independent optimization strategies based on the dual-parameter synergistic inhibition effect, and generates an adaptive quality control instruction sequence that includes response time synchronization strategy, execution status feedback code and anomaly handling jump code after equipment technical specification feasibility verification and safety verification.

[0015] According to another aspect of the present invention, a quality monitoring system for the production process of modified asphalt is provided, comprising: The collection module is used to acquire temperature data and stirring torque data at multiple points in the modified asphalt melting and blending tank. Combined with the real-time angle position of the stirring blades, a spatiotemporal coupled dataset of temperature and torque is constructed through a spatiotemporal alignment mechanism. The extraction module is used to extract the time series of torque data and the fluctuation curve of temperature with blade phase from the temperature-torque spatiotemporal coupling dataset. After being subjected to fast Fourier transform, the data is coupled in the frequency domain to generate a global coupled energy distribution vector. The key feature parameters are extracted and weighted to obtain the feature vector of the modifier dispersion state. The prediction module is used to map the feature vector of the dispersion state of the modifier to the phase space coordinate system, identify the system state transition nodes through the topological changes of the phase space orbit, and construct a mass state transition prediction function. The adjustment module is used to obtain the optimal combination of stirring parameters by optimizing and adjusting the parameters based on the output of the mass state transition prediction function, and to generate an adaptive mass control command.

[0016] Compared with existing technologies, this invention constructs a three-dimensional spatiotemporal coupled dataset of temperature, torque, and phase by introducing the real-time angular position of the stirring blades, and generates a feature vector of the modifier dispersion state through frequency domain coherence coupling analysis. This enables online characterization of the dispersion state differences caused by shear nonuniformity, overcoming the shortcomings of traditional methods that rely solely on time-domain averages and cannot identify bimodal particle size distribution and softening point fluctuations. On this basis, the invention predicts the probability of quality state transitions through phase space trajectory analysis and three-sub-model fusion, and uses a two-parameter iterative optimization based on parameter sensitivity and reverse step size allocation, combined with a response time difference synchronous execution strategy, to achieve an upgrade from passive hysteresis detection to active forward-looking control, significantly improving the quality stability and batch consistency of modified asphalt continuous production. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart of a quality monitoring method in the production process of modified asphalt according to an embodiment of the present invention. Detailed Implementation

[0018] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention; it should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0019] As mentioned in the background section above, the existing technology has two main problems: First, the dynamic coupling characteristics of temperature and torque during the modified asphalt production process have not been fully captured and analyzed, resulting in an incomplete assessment of the uniformity of modifier dispersion in the molten state, making it difficult to accurately reflect the global characteristics of the system; Second, there is a lack of a real-time quality control mechanism based on system state transitions, and the process optimization method relying on single parameter feedback has limited adjustment accuracy, making it difficult to meet the dual requirements of high efficiency and high precision in modern production.

[0020] Figure 1 This is a system block diagram of a quality monitoring method for modified asphalt production according to an embodiment of the present invention. Figure 1 As shown, the quality control methods in the modified asphalt production process include: S1: Obtain multi-point temperature data and stirring torque data inside the modified asphalt melting and blending tank, and construct a temperature-torque spatiotemporal coupled dataset by combining the real-time angle position of the stirring blades through a spatiotemporal alignment mechanism.

[0021] Multiple sets of high-temperature resistant temperature sensors are distributed circumferentially and axially along the outer wall and internal guide tube of the melt-blending tank in the modified asphalt production line. The operating temperature range of these sensors covers 120℃ to 220℃, enabling stable acquisition of instantaneous temperature values ​​of the asphalt matrix at different spatial locations within the tank during the melt-blending process. Simultaneously, a high-precision torque sensor is installed on the agitator spindle to collect the instantaneous torque values ​​experienced by the agitator during the melt-blending process. An angle encoder is also installed on the agitator spindle to record the current angular position of the agitator blades in real time.

[0022] Based on the synchronous acquisition of temperature sensors, torque sensors, and angle encoders, a spatiotemporally coupled temperature-torque dataset is constructed. Specifically, a synchronous trigger signal ensures that each sensor samples at the same time reference, and the multiple temperature values, torque values, and blade phase angles at each sampling moment are combined into a sampling tuple. A spatiotemporal alignment mechanism based on blade phase binds the instantaneous temperature value of each temperature acquisition point to the current angular position of the stirrer blade, as expressed by the formula:

[0023] in, This is a temperature-torque spatiotemporal coupling mapping. For the first At time 1, temperature measuring points Temperature value, For the stirrer blades at any time phase angle, The angular velocity of the stirrer. This represents the total number of temperature measurement points.

[0024] Furthermore, to address potential sensor noise and transient interference during the sampling process, the original temperature and torque signals are preprocessed using a sliding window mean filter. The filter window length is determined based on the stirrer's rotation cycle to ensure that the effective fluctuation characteristics related to the stirring cycle are not disrupted. If data for a certain temperature measurement point is missing in multiple consecutive sampling cycles, it is filled in using bilinear interpolation of adjacent measurement points and historical data to construct a complete spatiotemporal coupled temperature and torque dataset.

[0025] The temperature-torque spatiotemporal coupling dataset is organized in a three-dimensional data structure. The first dimension is indexed according to the spatial distribution of temperature sensors in the molten blending tank, with each temperature measurement point corresponding to a unique spatial index identifier. The second dimension is periodically arranged according to the angular position of the stirrer blades, dividing the complete rotation cycle of the stirrer into several angular intervals and assigning corresponding phase indices. The third dimension is arranged according to the time sequence of data acquisition, with each sampling moment corresponding to a time step index.

[0026] If it is necessary to store data at a specific time and location, the corresponding data storage location is located by combining three-dimensional indexes. When the data acquisition system obtains new temperature and torque measurements, it stores the data in the corresponding location of the three-dimensional data structure according to the current spatial location index, blade phase index, and time step index. Each data tuple in the storage location contains the temperature measurement value, torque measurement value, and related location identification information for that spatiotemporal point.

[0027] S2: Extract the time series of torque data and the temperature fluctuation curve with blade phase from the temperature-torque spatiotemporal coupling dataset, perform frequency domain coupling after fast Fourier transform, generate a global coupling energy distribution vector, extract key feature parameters and weighted combination, and obtain the feature vector of the modifier dispersion state.

[0028] From the acquired three-dimensional data structure of the temperature-torque spatiotemporal coupled dataset, using fixed spatial and phase indices as arbitrary values, and varying only the time step index, incrementing sequentially from the first time step, the storage location corresponding to each time step in the three-dimensional data structure is accessed one by one. The torque measurement value field within the data tuple at that storage location is read, skipping the temperature measurement value and location identification information fields within the same data tuple. After reading the torque measurement value of the first time step, the time step index is incremented by one unit, and the process continues to read the torque measurement value from the storage location of the next time step, repeating this process until all storage locations corresponding to all time step indices have been traversed. Once the torque data for all time steps has been read, the read torque measurement values ​​are arranged sequentially in a one-dimensional array according to their time step index, forming a time series of stirring torque data with time as the horizontal axis and torque measurement values ​​as the vertical axis.

[0029] After obtaining the stirring torque data time series, the number of data points in the time series is checked using binary representation. If the number of data points is not equal to an integer power of two, data points with a value of zero are added to the end of the time series so that the total number of data points reaches a value greater than the least power of two of the original data length.

[0030] After the data length is adjusted, the fast Fourier transform algorithm is used to divide the adjusted time series into two subsequences according to the odd and even positions. Each subsequence is then grouped according to the odd and even positions, and this grouping process is repeated until each subsequence contains only one data point. After grouping, starting from the smallest single-point subsequence, calculate the complex multiplication and addition operations between two adjacent subsequences, and merge the calculations upwards level by level until the Fourier transform result of the entire time series is obtained.

[0031] Furthermore, during the melt blending process of modified asphalt, the original Fast Fourier Transform (FFT) results cannot directly distinguish which frequency components truly correspond to changes in the dispersion state of the modifier. To accurately extract the frequency characteristics related to the dispersion state of the modifier from the FFT results of the torque signal and suppress equipment interference noise, each frequency component in the transform results undergoes targeted correction processing. The corrected amplitude value... :

[0032] in, This is the corrected amplitude value. For frequency index, The normalization coefficient for the spectral amplitude is... The magnitude of the original Fast Fourier Transform result. The Gaussian window attenuation intensity parameter, This is the frequency index corresponding to the characteristic frequency of the stirrer. This is the frequency window width parameter. This is the frequency offset suppression coefficient. This is the frequency index corresponding to the basic operating frequency of the mixer.

[0033] After applying the correction formula to each frequency component in the Fourier transform result, a series of corrected amplitudes are obtained. , , etc.

[0034] After calculating the corrected amplitudes for all frequency components, the phase angle information of each frequency component in the original Fourier transform result is retained unchanged, and the corrected amplitudes are... With the corresponding original phase angle Recombined into a complex form, the frequency domain feature representation, i.e. .

[0035] Complex frequency domain eigenvalues ​​of all frequency components in ascending order of frequency index, starting from zero. The phase angle information is arranged sequentially in an array. Each element of the array contains the corrected amplitude and phase angle of a frequency component. The entire array constitutes the frequency domain characteristic spectrum of torque data optimized for the dispersion state of the modifier, in which key frequency components related to the dispersion process of the modifier are enhanced, while noise frequency components are suppressed.

[0036] Based on the frequency domain feature spectrum of torque data containing amplitude and phase information, frequency domain information on temperature fluctuations with blade phase is further extracted from the temperature-torque spatiotemporal coupling dataset. Specifically: With a fixed spatial index in the 3D data structure, traverse all phase indices and time step indices for each spatial measurement point. The temperature values ​​of the measuring point at different times and when the blade is in a specific phase interval are extracted in ascending order of phase index. A two-dimensional matrix is ​​constructed, where rows correspond to time steps and columns correspond to phase intervals. The arithmetic mean of the temperature values ​​in each column (i.e., the same phase interval) is calculated in the time dimension to obtain the average fluctuation curve of temperature at the measuring point as a function of blade phase within a complete stirring cycle, denoted as . ,in For discretized phase index; For the average fluctuation curve Using the same Fast Fourier Transform (FFT) processing procedure as the torque signal, the temperature-phase fluctuation frequency domain characteristic spectrum of this spatial measurement point is obtained, denoted as... It is also a complex array containing the correction amplitude and phase angle of each frequency component.

[0037] Repeat the above process to traverse all spatial measurement points and obtain the temperature-phase fluctuation frequency domain characteristic spectrum of each measurement point. Subsequently, the frequency domain characteristic spectrum of the torque data was analyzed. The temperature-phase fluctuation frequency domain characteristic spectrum of each spatial measurement point was compared with that of the other points. Coherent coupling operations are performed in the frequency domain, with the coupling method being: at the same frequency index. Next, calculate the conjugate product of the two spectra, expressed by the formula:

[0038] in, This indicates taking the conjugate complex number.

[0039] right Take the modulus to get the first Temperature-torque frequency domain coupled amplitude spectrum at each spatial measurement point , This characterizes the temperature fluctuations and torque fluctuations at this spatial location at different frequencies. The higher the common energy density at that frequency, the tighter the thermo-mechanical coupling and the higher the correlation with the dispersion process of the modifier under blade shear.

[0040] Coupled amplitude spectrum for each spatial measurement point Peak detection and frequency band division are performed: traversal For all frequency components, the amplitude of each component is squared to obtain its energy value. The main frequency components are determined by sorting them in descending order of energy value. The midpoint between two adjacent main frequency component positions is used as the boundary point of the frequency band. The entire spectrum from zero frequency to the highest frequency is divided into continuous frequency band intervals. The energy values ​​of all frequency components in each frequency band interval are summed to obtain the main frequency band energy distribution vector of the spatial measurement point.

[0041] Arrange the energy distribution vectors of the main frequency bands of all spatial measurement points in spatial index order to form a two-dimensional energy distribution matrix, where rows correspond to spatial locations and columns correspond to frequency bands. Calculate the arithmetic mean of each column of this matrix (i.e., the energy values ​​of the same frequency band at all spatial locations) to obtain a one-dimensional vector with the same number of frequency bands. This vector reflects the average distribution of temperature-torque frequency domain coupling energy across different frequency bands within the entire blending tank space, and is denoted as the global coupling energy distribution vector.

[0042] Three key parameters—peak position, peak amplitude, and bandwidth—are extracted from the aforementioned global coupled energy distribution vector. Specifically, the frequency band with the highest energy value is detected from the global coupled energy distribution vector and designated as the peak position; the energy value corresponding to this frequency band is designated as the peak amplitude; and the sum of the bandwidths of all consecutive frequency bands with energy values ​​exceeding half of the peak amplitude is designated as the bandwidth. These three key parameters are then linearly combined according to predefined weighting coefficients to generate a modifier dispersion state feature vector that characterizes the dispersion state of the modifier during melt blending.

[0043] The formula for calculating the weighting coefficient is as follows:

[0044] in, For the first Predefined weight coefficients corresponding to each feature quantity The first one obtained by the least squares method The regression coefficients of the characteristic quantities For the first Peak amplitude of each sample The maximum peak amplitude among all samples is used for normalization. For the first The convolution peak positions of each sample The median of the peak positions of all samples is used as the reference peak position. The standard deviation of the peak position. For the first The bandwidth of the frequency distribution of each sample The median of the bandwidths of all samples is used as the reference bandwidth. The minimum bandwidth threshold, The total number of samples, This is the regularization parameter.

[0045] S3: Map the feature vector of the dispersion state of the modifier to the phase space coordinate system, identify the system state transition nodes through the topological changes of the phase space orbit, and construct a mass state transition prediction function.

[0046] Based on the obtained feature vector sequence of the modifier dispersion state, before performing phase space mapping, the feature vector values ​​and timestamps are first normalized to eliminate the influence of dimensional differences on orbital geometry analysis. Specifically, the feature vector values ​​at all times are linearly mapped to the [0,1] interval using a minimum-maximum normalization method. At the same time, the timestamp values ​​are converted into relative time offsets with the first sampling time as the zero point, and the same normalization process is performed on the relative time offsets to obtain a dimensionless feature value sequence and a dimensionless time series.

[0047] Create a three-dimensional Cartesian coordinate system as the phase space coordinate system, read the characteristic vector value of the modifier dispersion state acquired at the first time step, and directly assign this value to the first coordinate axis of the three-dimensional coordinate system. The coordinate value of the axis is obtained by reading the timestamp value corresponding to the feature vector and directly assigning the timestamp value to the second coordinate axis of the three-dimensional coordinate system. The coordinate values ​​of the axis, for the first time step, since there is no data from the previous time step for comparison, will... Set the axis coordinate values ​​to zero; After generating the three-dimensional coordinates at the first time step, the eigenvector values ​​of the modifier dispersion state at the second time step are read as... The axis coordinate value is read from the timestamp value of the second moment. The axis coordinate value is calculated by subtracting the eigenvector value at the first time step from the eigenvector value at the second time step. The feature vector values ​​at each time step are processed sequentially using the same method, from the third and fourth time steps to the last time step. The feature vector value at each time step is then used as... Axis coordinates, timestamp as The axis coordinates are the difference between the feature vectors at the current time and the previous time. Axis coordinates. After calculating the 3D coordinates for all time points, these coordinates are stored in an array structure in ascending order of timestamp, with each element of the array containing a 3D coordinate point. , , The coordinate values ​​and their corresponding time markers form a complete sequence of phase space coordinate points.

[0048] Based on the acquired phase space coordinate point sequence, the three-dimensional coordinate value of the first coordinate point is read from the sequence as the starting point of the orbit curve. Then, the three-dimensional coordinate value of the second coordinate point is read, and the first and second coordinate points are connected by straight line segments in three-dimensional phase space. All adjacent coordinate point pairs are connected sequentially in chronological order to form a continuous broken-line orbit composed of multiple straight line segments connected end to end.

[0049] The algorithm iterates through each straight line segment in the track curve, calculating the angle between the direction vectors of two adjacent line segments as the track's turning angle. Simultaneously, for each internal coordinate point on the track (excluding the beginning and end points), it calculates the discrete curvature at that point using the two line segments formed by its two preceding and following coordinate points. Curvature is defined as the sine of the supplementary angle formed by two adjacent line segments with the current point as the vertex, multiplied by the geometric mean of the lengths of the two line segments, and scaled. A larger value indicates a more severe curvature at that point. The curvature of all internal points is calculated sequentially, forming a curvature sequence.

[0050] The severity of system state changes and key transition nodes are identified by analyzing the numerical distribution patterns of turning angles and curvatures. The specific rules are as follows: If the direction vectors of all straight line segments of the orbit curve are... All axis components are positive and When all axial components are positive, it indicates that the characteristic value of the modifier dispersion state increases monotonically with time, and at this time it is in a stable evolution stage of continuous improvement. If multiple inflection points with turning angles greater than 90 degrees appear consecutively in the trajectory curve, and these inflection points are evenly distributed on the time axis, it reflects that there is periodic oscillation in the dispersion process of the modifier, and the oscillation period corresponds to the time interval between adjacent large-angle inflection points. If the trajectory curve forms a closed or nearly closed loop structure in three-dimensional space, that is, when the spatial distance between the coordinates of the end point and the coordinates of the start point is less than one-tenth of the total length of the trajectory, it indicates that the trajectory is cyclically moving within the spatial region of the state, corresponding to the periodic regression phenomenon of the dispersion state of the modifier. If the direction vector of a continuous line segment in the trajectory curve suddenly changes in the opposite direction, that is, the angle between the direction vectors of the previous line segment and the next line segment is close to 180 degrees, it indicates that a state transition has occurred, and the dispersion state of the modifier has been drastically reversed at that moment. If the distribution range of the last few coordinate points of the trajectory curve in three-dimensional space is less than one-twentieth of the distribution range of the entire trajectory, and the length of the line connecting these points tends to zero, it indicates that the curve has converged to a stable state and the modifier dispersion process has reached the final equilibrium state.

[0051] Furthermore, starting from the second element of the turning angle array, each element is checked sequentially, comparing its size with the preceding and following elements. If the current element is greater than both the preceding and following elements, its array index is recorded as a local maximum point of the turning angle. The same comparison method is used to traverse the curvature array, identifying local maxima points that are simultaneously greater than their adjacent elements, and recording their array index positions. The absolute value of the difference between the index position of each turning angle maximum point and the index position of each curvature maximum point is calculated. If the absolute value of the difference between the index positions of a turning angle maximum point and a curvature maximum point is less than a preset proximity range, the corresponding time position is determined as a system state transition node.

[0052] Simultaneously, the trajectory pattern recognition results are transformed into auxiliary positioning conditions: if a certain time period is identified as "direction vector reversal" (state transition), the start and end points of the reversal line segment are directly added as candidate state transition nodes; if a "closed loop" pattern appears, the closure point of the loop is also marked as a candidate node. The candidate nodes obtained from quantitative extreme value detection are merged with the candidate nodes supplemented by qualitative patterns, and the union of the time indices is taken as the final set of system state transition nodes, and the time position corresponding to each node is recorded.

[0053] Based on the identified state transition node locations and their corresponding time information, a correlation model is established between the occurrence time of state transition nodes and the dispersion quality state of the modifier, including a state transition node time correlation sub-model, a track geometric feature correlation sub-model, and a historical quality state sequence correlation sub-model.

[0054] Once the time location information of the state transition node is obtained, the time-related sub-model First, Gaussian weights are assigned to each transition node, and the temporal distance similarity is calculated with the reference time base point as the center. Simultaneously, a unit step function is used... The time intervals between adjacent nodes are processed to ensure that the causal relationship of the time series is preserved, and a standardized correlation degree value reflecting the time evolution pattern is output, expressed by the formula:

[0055] in, For the first The time position of each state transition node. As a reference time base point, The standard deviation parameter for time location. Adjacent state transition nodes and The time interval between For unit step function, This represents the total number of state transition nodes.

[0056] Geometric feature correlation sub-model The system receives the turning angle and curvature data from each node on the track curve. It then uses the arctangent function to calculate the ratio of the cosine-weighted sum of the angle information to the sine-weighted sum of the curvature information, and multiplies this ratio by the root mean square value of the curvature. This results in a correlation function value that comprehensively reflects the geometric changes of the track. The formula is as follows:

[0057] in, For the first The turning angle of the trajectory curve at each node For the first The curvature values ​​at each node. This represents the number of sampling points for geometric features.

[0058] Historical state sequence association sub-model The historical quality status levels are subjected to exponential decay weighting. It's important to clarify that the "quality status level" here refers to the actual quality index of modified asphalt obtained from offline sampling and testing (such as softening point temperature testing), synchronized with the online feature vector sequence via timestamp alignment. Specifically, each offline test yields a quality status level value, which is then matched with the most recent time point in the phase space orbit based on the testing sampling time, forming a discrete historical quality status sequence. . Through attenuation factor It reflects the time decay effect of historical information, and introduces a quality state change stationarity correction term to suppress the excessive influence of historical fluctuations on current predictions, outputting a quantitative representation of the historical evolution trend, expressed by the formula:

[0059] in, For the first time in history The quality status level at any given moment. The historical state decay factor, and These are the maximum and minimum values ​​for the quality status level, respectively. The length of the historical state sequence.

[0060] Furthermore, the outputs of the three sub-models are linearly combined according to the normalized weight coefficients, and a logarithmic time correction term is added to capture the long-term evolution trend, thus obtaining the predicted quality state level of the modifier dispersion. The formula is expressed as:

[0061] in, , , These are the weight coefficients for the three sub-models. This is a time correction factor. For predicting time intervals, This is a time-scale parameter.

[0062] Furthermore, the formula for calculating the weighting coefficient is as follows:

[0063]

[0064] in, For the first The weight coefficients of each sub-model For the first The correlation coefficient between the sub-model and the actual quality state level of the modifier dispersion. For the first The sub-model for the first The calculated value of the nth historical sample, \bar{F_j} is the nth The arithmetic mean of the historical calculated values ​​of each sub-model For the first The actual modifier dispersion quality status level of a historical sample. This is the arithmetic mean of the historical actual quality status levels. This represents the total number of historical samples.

[0065] It should be noted that when historical data is lacking during the initial operation phase of the system, the weighting coefficients are set using an initialization strategy based on engineering experience and theoretical analysis. Once enough actual quality status data has been accumulated during operation, the weighting coefficients are dynamically adjusted and optimized using the correlation coefficient calculation formula, gradually replacing the initial experience-based settings.

[0066] Furthermore, a quality state transition prediction function is constructed, which integrates the quality state levels output by the comprehensive correlation model. As the basic input, the relationship between the quality state level and the transition threshold is evaluated, and the probability value of the quality state transition is output, which is expressed by the formula:

[0067] in, Let be the probability of a mass state transition. For transition sensitivity parameters, To predict the dispersion quality state grade of the modifier, This represents the threshold for quality state transitions. The cumulative distribution function of the standard normal distribution. For orbital geometric characteristics correlation function, and These are the mean and standard deviation parameters of the geometric features, respectively.

[0068] The mass state transition prediction function can predict the timing and direction of possible mass state abrupt changes during the dispersion of modifiers based on the evolution trend of the current phase space orbit and historical state transition patterns, providing forward-looking early warning information for the quality control of the melt blending process.

[0069] S4: Based on the output of the mass state transition prediction function, the optimal combination of stirring parameters is obtained through parameter optimization and adjustment, and an adaptive quality control command is generated.

[0070] The transition probability value output by the transition probability prediction function is used as the target constraint condition for the parameter optimization and adjustment process. The expected transition probability target value is set as the minimum risk level required for stable system operation.

[0071] Starting from the current transition probability value, parameter optimization is performed along the direction of reducing transition risk. All records are extracted from the historical database. Each record contains the stirring speed value S, stirring time value T, stirring power value P, stirring blade angle value A, stirring interval time value I, and the corresponding transition probability value R. If the number of records in the historical database is insufficient to support reliable statistical analysis (e.g., less than the preset minimum sample size threshold), a cold start strategy based on a mechanistic model is activated: based on the equipment nameplate parameters and process design values, combined with operational experience, standard operating points and safety boundaries for each parameter are preset. Using the standard operating points as the center and the direction of reducing transition probability as the guide, rule-driven trial steps are used for initial adjustment. At the same time, the parameter combination and corresponding transition probability feedback value of each trial are added to the historical database in real time until the accumulated data volume meets the requirements of statistical analysis, and then the system automatically switches to data-driven parameter optimization mode.

[0072] Single-parameter analysis was performed using stirring speed. Stirring time, stirring power, stirring blade angle, and stirring interval time were fixed as their respective average values. Only the stirring speed value was changed, gradually increasing from the historical minimum to the maximum value. The corresponding transition probability value was recorded for each fixed step. The change in transition probability of stirring speed was obtained by subtracting the minimum transition probability value from the maximum transition probability value. The same method was used to analyze stirring time, stirring power, stirring blade angle, and stirring interval time. The other four parameters were fixed as average values, and only the target parameter was changed from the minimum to the maximum value, recording the change in transition probability.

[0073] The transition probability changes of the five parameters are compared. The parameter with the largest change is designated as the first sensitive parameter, and the parameter with the second largest change is designated as the second sensitive parameter. Simultaneously, the values ​​of the first and second sensitive parameters are changed, while the other three parameters are fixed at their average values. Starting from the minimum combination of the two parameters, the values ​​are gradually increased to the maximum combination, and the maximum change in transition probability throughout the process is recorded as the dual-parameter change amplitude. The dual-parameter change amplitude is compared with the sum of the individual changes of the first and second sensitive parameters. If the dual-parameter change amplitude is less than the sum of the two individual parameter changes, these two parameters are identified as a key stirring parameter combination with a synergistic inhibition effect, and subsequent dual-parameter joint optimization is directly performed. If the dual-parameter change amplitude is equal to or greater than the sum of the two individual parameter changes, it indicates that there is no synergistic inhibition effect between the two parameters. Therefore, the dual-parameter joint optimization strategy is abandoned, and the optimization strategy degenerates into independent optimization—single-variable optimal value searches are performed on the first and second sensitive parameters within their respective value ranges, and the combination of the two independently searched optimal values ​​is output as the key stirring parameter combination.

[0074] After determining the key mixing parameter combination, read the value of the first sensitive parameter of the current system and record it as follows. The value of the second sensitive parameter is denoted as As the starting point for adjustment, the set transition probability target value is read and recorded as follows: ; Furthermore, the adjustment step size is determined based on the principle that parameter sensitivity and step size should have an inverse relationship: the change amplitude of the transition probability of the first sensitive parameter is compared with that of the second sensitive parameter. The parameter with a larger change amplitude has a more significant impact on the transition probability and should be assigned a smaller adjustment step size to achieve fine control. Conversely, the parameter with a smaller change amplitude can be assigned a relatively larger step size to accelerate the search speed. Specifically, the change amplitudes of the transition probabilities of the two parameters are normalized so that the sum of the two step sizes equals twice the base step size value, and the step size of each parameter is inversely proportional to its change amplitude.

[0075] The adjustment process reads the current transition probability value and records it as follows: ,calculate minus Obtain the error value ,like If the value is positive, both parameters are adjusted in the direction of reducing the transition probability. If the value is negative, both parameters will be adjusted in the direction of increasing the transition probability. The specific adjustment direction is determined based on the correlation between parameter increases / decreases and transition probability changes in historical data. The first round of adjustment process will Adjusting the step size by adding or subtracting the first sensitive parameter yields the result. ,Will Adjusting the step size by adding or subtracting the second sensitive parameter yields the result. ,Will and The new transition probability value obtained by inputting the transition probability prediction function is denoted as . ,calculate and The difference is used to obtain a new error value. ; Once the new error value is calculated, the adjustment process is checked. The relationship between the absolute value of the error and the preset convergence criterion; to enhance the robustness and convergence of the optimization process, the following auxiliary mechanisms are introduced: A maximum iteration limit is set; if the convergence criterion is not met even after reaching the maximum iteration limit, the iteration is terminated and the current optimal parameters are output, along with a message indicating incomplete convergence; a step size decay strategy is introduced; when the sign of the error value alternates repeatedly in several consecutive iterations (i.e., oscillation occurs), the current step size is reduced to half of the original step size, and the adjustment process is restarted. If the absolute value of the convergence criterion is greater than the convergence criterion, then... Assign to ,Will Assign to Repeat the above adjustment process if If the absolute value of the convergence criterion is less than or equal to the absolute value of the convergence criterion, then and Output as the optimal combination of key stirring parameters.

[0076] Based on the optimal combination of key mixing parameters, the minimum allowable value of the first sensitive parameter is read from the equipment technical specification database and denoted as follows: And the maximum allowed value is denoted as Simultaneously, the minimum allowable value of the second sensitive parameter is read and denoted as... And the maximum allowed value is denoted as ; The value of the first sensitive parameter in the optimal parameter combination and If a comparison is made, Less than Then Reassigned ,like Greater than Then Reassigned ,like exist and Maintaining between The value remains unchanged, and the same comparison and assignment logic is used to process the value of the second sensitive parameter. .

[0077] Further, the current device temperature value is read from the system monitoring database and recorded as follows: The current vibration amplitude value is recorded as The current motor load value is recorded as follows: Simultaneously, the upper limit of equipment temperature safety is read from the historical safe operation database and recorded as follows: The upper limit of the vibration amplitude is denoted as The upper limit of motor load safety is denoted as ; Calculate the expected equipment operating indicators based on the parameter adjustment range, and then... Subtract the current value of the first sensitive parameter to obtain the adjustment amount of the first parameter. ,Will Subtract the current value of the second sensitive parameter to obtain the adjustment amount of the second parameter. Calculated using a pre-calibrated device response model and Corresponding equipment temperature increment Increment of vibration amplitude and motor load increment ; Furthermore, Plus Obtain the expected equipment temperature ,Will Plus Obtain the expected vibration amplitude ,Will Plus Get the expected motor load Check them separately Does it exceed , Does it exceed , Does it exceed If any expected indicator exceeds the corresponding safety limit, the scale will be reduced proportionally. and Until all expected indicators do not exceed the safety limit, the reduced adjustment amount is recalculated to obtain the final executable first and second sensitive parameter values.

[0078] The optimal combination of stirring parameters, verified for feasibility and safety, is retrieved from the equipment status database, and the actual value of the current primary sensitive parameter is recorded as follows: The actual value of the second sensitive parameter is denoted as The target value of the first sensitive parameter in the optimal parameter combination. minus Obtain the first parameter adjustment amount The target value of the second sensitive parameter minus Obtain the adjustment amount of the second parameter ; According to the equipment control protocol Converted into control commands, if When the value is positive, a parameter increment instruction is generated. The instruction format is the device address code of the first sensitive parameter plus the increment opcode plus... The absolute value, if When the value is negative, a parameter reduction instruction is generated. The instruction format is the device address code of the first sensitive parameter plus the reduction opcode plus... The absolute value is processed using the same judgment and transformation logic. Control instructions for generating the second sensitive parameter; Furthermore, the response time for reading the first sensitive parameter from the equipment's technical specifications. Response time of the second sensitive parameter To ensure that the effects of adjusting the two parameters take effect synchronously in time, a strategy of prioritizing the parameter with the longer response time is adopted: [The strategy is as follows:] and If a comparison is made, Greater than If the first sensitive parameter responds slowly, its control command is set to execute immediately; if the second sensitive parameter's control command is set to execute with a delay, the delay time is [not specified]. minus ;like Greater than If the second sensitive parameter responds more slowly, its control command is set to execute immediately, while the control command for the first sensitive parameter is set to execute with a delay of [time value missing]. minus .like and If they are equal, the two instructions are sent and executed simultaneously.

[0079] The two parameter control instructions are arranged in the order of execution time. An execution status feedback code is added after each instruction to confirm whether the instruction was executed successfully. An exception handling jump code is also added so that when the instruction fails to execute, it will automatically jump to the backup parameter adjustment scheme.

[0080] Meanwhile, a quality monitoring trigger condition judgment instruction is inserted at the beginning of the instruction set. This instruction continuously monitors the change in the transition probability value. If the absolute value of the difference between the transition probability value and the previous recorded value exceeds the preset change amount, the parameter optimization and adjustment process is automatically called back and a new control instruction sequence is generated to realize real-time response and automatic control of quality status changes.

[0081] In summary, the quality monitoring method for modified asphalt production based on the embodiments of the present invention has been clarified. It constructs a three-dimensional spatiotemporal coupled dataset of temperature, torque, and phase by introducing the real-time angular position of the stirring blades, and generates a feature vector of the modifier dispersion state through frequency domain coherence coupling analysis. This achieves online characterization of the dispersion state differences caused by shear nonuniformity, overcoming the shortcomings of traditional methods that rely solely on time-domain averages and cannot identify bimodal particle size distributions and softening point fluctuations. Furthermore, it predicts the probability of quality state transitions through phase space trajectory analysis and three-sub-model fusion, and uses a two-parameter iterative optimization based on parameter sensitivity and reverse step size allocation, combined with a response time difference synchronous execution strategy. This achieves an upgrade from passive lag detection to proactive forward control, significantly improving the quality stability and batch consistency of continuous modified asphalt production.

Claims

1. A quality monitoring method in the production process of modified asphalt, characterized in that, include: Multi-point temperature data and stirring torque data inside the modified asphalt melt blending tank were acquired, and combined with the real-time angle position of the stirring blades, a spatiotemporal coupled dataset of temperature and torque was constructed through a spatiotemporal alignment mechanism. The time series of torque data and the fluctuation curve of temperature with blade phase are extracted from the temperature-torque spatiotemporal coupling dataset. After being subjected to fast Fourier transform, they are coupled in the frequency domain to generate a global coupled energy distribution vector. The key feature parameters are extracted and weighted to obtain the feature vector of the modifier dispersion state. The feature vector of the dispersion state of the modifier is mapped to the phase space coordinate system, and the system state transition nodes are identified by the topological changes of the phase space orbits, thereby constructing a mass state transition prediction function. Based on the output of the mass state transition prediction function, the optimal combination of stirring parameters is obtained through parameter optimization and adjustment, and an adaptive quality control command is generated.

2. The quality monitoring method in the production process of modified asphalt according to claim 1, characterized in that, The spatiotemporal alignment mechanism binds the instantaneous temperature values ​​of each temperature acquisition point to the current angular position of the stirrer blades based on the blade phase, constructing a three-dimensional data structure. The first dimension is indexed according to the spatial distribution of temperature sensors, the second dimension is arranged periodically according to the angular position of the stirrer blades, and the third dimension is arranged according to the data acquisition time sequence.

3. The quality monitoring method in the production process of modified asphalt according to claim 1, characterized in that, The Fast Fourier Transform includes a correction process for the original transform result, and the correction formula is expressed as follows: ； in, This is the corrected amplitude value. For frequency index, The normalization coefficient for the spectral amplitude is... The magnitude of the original Fast Fourier Transform result. The Gaussian window attenuation intensity parameter, This is the frequency index corresponding to the characteristic frequency of the stirrer. is the frequency window width parameter, is the frequency offset suppression coefficient, and is the frequency index corresponding to the basic operating frequency of the stirrer.

4. The quality monitoring method in the production process of modified asphalt according to claim 3, characterized in that, The frequency domain coupling is achieved by calculating the conjugate product of the frequency domain characteristic spectrum of the torque data and the frequency domain characteristic spectrum of the temperature-phase fluctuation, as expressed by the formula: ； in, This indicates taking the conjugate complex number. The frequency domain characteristic spectrum of temperature-phase fluctuation at the spatial measurement point. This is the frequency domain characteristic spectrum of torque data.

5. The quality monitoring method in the production process of modified asphalt according to claim 4, characterized in that, The phase space coordinate system is a three-dimensional rectangular coordinate system, and the eigenvector values ​​are used as... Axis coordinates, timestamp as The axis coordinates, the difference between the feature vectors of the current time and the previous time, are used as... The system state transition nodes are determined by combining the detection of local maxima of track turning angles and curvature with track pattern recognition.

6. The quality monitoring method in the production process of modified asphalt according to claim 5, characterized in that, The quality state transition prediction function includes a time-related sub-model, a geometric feature-related sub-model, and a historical state sequence-related sub-model. It outputs the probability value of a quality state transition by linearly combining dynamic weight coefficients and adding a logarithmic time correction term. The formula is as follows: ； in, Let be the probability of a mass state transition. For transition sensitivity parameters, To predict the quality state grade of the modifier dispersion, This represents the threshold for quality state transitions. The cumulative distribution function of the standard normal distribution. For orbital geometric characteristics correlation function, and These are the mean and standard deviation parameters of the geometric features, respectively.

7. The quality monitoring method in the production process of modified asphalt according to claim 6, characterized in that, The parameter optimization adjustment determines the first and second sensitive parameters through single-parameter sensitivity analysis. Based on the dual-parameter synergistic inhibition effect, it verifies and selects a joint optimization or independent optimization strategy. After the feasibility and safety verification of the equipment technical specifications, it generates an adaptive quality control instruction sequence that includes a response time synchronization strategy, execution status feedback code, and anomaly handling jump code.

8. A quality monitoring system for the production process of modified asphalt, characterized in that, include: The collection module is used to acquire temperature data and stirring torque data at multiple points in the modified asphalt melting and blending tank. Combined with the real-time angle position of the stirring blades, a spatiotemporal coupled dataset of temperature and torque is constructed through a spatiotemporal alignment mechanism. The extraction module is used to extract the time series of torque data and the fluctuation curve of temperature with blade phase from the temperature-torque spatiotemporal coupling dataset. After being subjected to fast Fourier transform, the data is coupled in the frequency domain to generate a global coupled energy distribution vector. The key feature parameters are extracted and weighted to obtain the feature vector of the modifier dispersion state. The prediction module is used to map the feature vector of the dispersion state of the modifier to the phase space coordinate system, identify the system state transition nodes through the topological changes of the phase space orbit, and construct a mass state transition prediction function. The adjustment module is used to obtain the optimal combination of stirring parameters by optimizing and adjusting the parameters based on the output of the mass state transition prediction function, and to generate an adaptive mass control command.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the quality monitoring method in the production process of modified asphalt as described in any one of claims 1 to 7.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the quality monitoring method in the production process of modified asphalt as described in any one of claims 1 to 7.

Images