Method and system for recognizing batch-to-batch impurity difference of OLED material

Abstract

The application provides a multi-batch OLED material impurity difference identification method and system, relates to the technical field of material detection, and comprises the following steps: obtaining chromatographic detection data of multiple batches of OLED materials, and constructing an original chromatographic sequence; extracting peak shape structure characteristic parameters, constructing a corresponding shape enhancement sequence; constructing a cross-batch reference center spectrum, and generating an aligned chromatographic sequence; inputting a decomposition channel, constructing a pure impurity response signal, and performing normalization and local enhancement processing; constructing a multi-scale difference representation and establishing a dynamic control threshold, performing abnormal signal determination, and outputting a batch impurity difference identification result. The application solves the technical problems in the prior art that the real impurity difference between different batches is easily covered or misjudged due to chromatographic retention time drift, baseline interference and noise influence, and rigid determination standards, and improves the OLED material quality control accuracy.

Description

Technical Field

[0001] This application relates to the field of materials testing technology, specifically to a method and system for identifying impurity differences in multiple batches of OLED materials. Background Technology

[0002] In the preparation and quality control of OLED materials, trace impurities often exist between different production batches. These trace impurities have a decisive impact on the performance and reliability of the devices. In actual analysis, due to unavoidable fluctuations in instrument status and experimental conditions, retention time drift is common in chromatographic data from different batches. This leads to inconsistencies in the position of the same impurity peak in multiple batches of spectra. Direct comparison by superposition can easily produce false positives or mask true differences. At the same time, baseline drift and high-frequency noise interference further weaken the identification of trace impurity peaks, causing critical impurities with low content to be often submerged or misjudged, affecting the reliability of quality control conclusions.

[0003] In summary, existing technologies suffer from technical problems such as chromatographic retention time drift, baseline interference and noise, and rigid judgment criteria, which can easily mask or misjudge the true differences in impurities between different batches, further affecting the accuracy of OLED material quality control. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for identifying impurity differences in multiple batches of OLED materials, in order to solve the technical problem in the prior art that the true impurity differences between different batches are easily masked or misjudged due to chromatographic retention time drift, baseline interference and noise, and rigid judgment criteria, which further affects the accuracy of OLED material quality control.

[0005] To achieve the above objectives, this application provides a method and system for identifying impurity differences in multiple batches of OLED materials.

[0006] Firstly, this application provides a method for identifying impurity differences in multiple batches of OLED materials. This method is implemented using a system for identifying impurity differences in multiple batches of OLED materials. The method includes: acquiring chromatographic detection data from multiple batches of OLED materials; performing unified sampling processing on the chromatographic detection data; constructing an original chromatographic sequence with retention time as the sequence index and response intensity as the numerical characterization; and based on the original chromatographic sequence, extracting peak shape structural feature parameters from each batch of chromatographic data. These structural feature parameters include peak position, peak width, and peak shape variation trend. The method also utilizes these peak shape structural feature parameters to construct a system for identifying impurity differences in multiple batches of OLED materials. The morphological enhancement sequence is obtained; after fusing the morphological enhancement sequence and the original chromatographic sequence, a cross-batch reference center spectrum is constructed, and nonlinear alignment processing of the original chromatographic sequence is performed under the constraint of the cross-batch reference center spectrum to generate an aligned chromatographic sequence; the aligned chromatographic sequence is input into the decomposition channel, and decomposition processing of adaptive baseline fitting and residual signal separation is performed to construct a pure impurity response signal after removing baseline interference, and normalization and local enhancement processing are performed to form a cross-batch impurity feature sequence; a multi-scale difference characterization quantity is constructed based on the impurity feature sequence, and a dynamic control threshold is established by combining historical batch statistical characteristics. The abnormal signal judgment of the multi-scale difference characterization quantity is performed using the dynamic control threshold, and the batch impurity difference identification result is output.

[0007] Optionally, the first-order gradient sequence and the second-order curvature sequence are calculated along the retention time direction of the original chromatographic sequence, and the chromatographic signal is divided into candidate intervals according to the position of gradient sign change to determine the time boundary of each peak structure; within each peak candidate interval, the peak position is determined based on the maximum response pad of the peak region, and the peak width position is determined based on the time span corresponding to the signal attenuation to a preset ratio threshold on both sides of the peak position, and the peak shape asymmetry parameter is constructed according to the gradient integral difference on both sides of the peak position; the peak position, peak width position, peak shape asymmetry parameter and curvature change trend parameter are coupled and fused, and the original chromatographic sequence is locally weighted and reconstructed using the coupling and fusion result to obtain the morphology-enhanced sequence.

[0008] Optionally, the morphological enhancement sequences of each batch are weighted and fused with the corresponding original chromatographic sequences to form a fused sequence that highlights the peak shape and structural features. Based on the fused sequence, the corresponding peak positions and peak structure intervals of each batch are extracted, and a set of cross-batch peak structure correspondences is constructed by matching and aligning the same peak structures in different batches. Under the constraints of the set of peak structure correspondences, weighted superposition and position correction processing are performed on the fused sequences of each batch within the corresponding peak structure intervals, and smooth fusion is performed on non-peak structure intervals to generate a cross-batch reference center spectrum with uniform peak structure distribution features.

[0009] Optionally, a sliding analysis window is constructed along the retention time direction for the aligned chromatographic sequence. Within each sliding analysis window, a set of local minima of characteristic signal intensity is extracted, and an initial baseline control node sequence is generated based on these local minima. An initial baseline estimation curve is constructed by piecewise interpolation fitting of the initial baseline control node sequence. Baseline stripping is then performed on the aligned chromatographic sequence using the initial baseline estimation curve as a constrained envelope to obtain residual signals. Based on the continuity of the residual signals between adjacent sliding analysis windows, the rate of change of signal gradient and the rate of change of curvature are calculated to construct a peak shape characterization. The continuity discrimination index for structural stability is used. Based on the continuity discrimination index, structural components are screened from the residual signal. Signal segments that simultaneously satisfy the amplitude continuity constraint, gradient monotonic change constraint, and curvature consistency constraint are identified as structural response components. The remaining discrete high-frequency disturbance signals are removed as random noise components. Cross-batch scale uniform processing is performed on the structural response components. The amplitude of each batch of structural response components is normalized by the peak intensity ratio mapping relationship based on the cross-batch reference center spectrum. Signal contrast enhancement is performed in the local time neighborhood corresponding to each structural response component to form an impurity feature sequence.

[0010] Optionally, within each sliding analysis window, a first-order gradient sequence and a second-order curvature sequence are calculated based on the residual signal, and a continuity discrimination index is constructed by combining the signal amplitude change. The continuity discrimination index is used to characterize the amplitude continuity, gradient change direction consistency and curvature change stability of the signal in the local time neighborhood.

[0011] Optionally, using the continuity discrimination index as a constraint, the residual signal is divided into candidate structural segments, and signal segments whose continuity discrimination index remains stable within a preset interval are identified as candidate structural segments; amplitude continuity constraint verification is performed on the candidate structural segments to establish a first verification result; monotonicity interval division is performed on the gradient sequence of the candidate structural segments to establish a second verification result; the curvature consistency of the peak region of the candidate structural segments is verified based on the curvature sequence change trend to generate a third verification result; when the first verification result, the second verification result, and the third verification result are all passing results, a structural response component is established.

[0012] Optionally, cross-window consistency verification is performed on the structural response components within adjacent sliding analysis windows. By comparing the peak position offset, peak width change rate, and gradient direction consistency in adjacent sliding analysis windows, stable and drifting segments of the structural response components are identified. Local time axis correction is performed using the stability decay degree of the drifting segments, and cross-batch scale unification is performed based on the correction results and the stable segments.

[0013] Optionally, the multi-scale difference representation quantity in the statistical characteristics of historical batches is read, and a time series distribution model is constructed using the multi-scale difference representation quantity; the degree of statistical distribution drift between different time batches is evaluated using the time series distribution model, and dynamic adaptive correction of the control threshold is performed based on the evaluation results to establish a dynamic control threshold.

[0014] Optionally, a batch anomaly identifier can be configured based on the batch impurity difference identification result, and the batch anomaly identifier can be used to perform anomaly diversion management of multiple batches of OLED materials.

[0015] Secondly, this application also provides a multi-batch OLED material impurity difference identification system for performing the multi-batch OLED material impurity difference identification method as described in the first aspect. The multi-batch OLED material impurity difference identification system includes: a unified sampling processing module for acquiring chromatographic detection data of multiple batches of OLED materials, performing unified sampling processing of the chromatographic detection data, and constructing an original chromatographic sequence with retention time as the sequence index and response intensity as the numerical characterization; and a feature parameter extraction module for extracting peak shape structural feature parameters from each batch of chromatographic data based on the original chromatographic sequence. The structural feature parameters include peak position, peak width position, and peak shape change trend, and the corresponding morphology enhancement sequence is constructed using the peak shape structural feature parameters. The reference center spectrum construction module is used to construct a cross-batch reference center spectrum after fusing the morphological enhancement sequence and the original chromatographic sequence. Under the constraint of the cross-batch reference center spectrum, it performs nonlinear alignment processing of the original chromatographic sequence to generate aligned chromatographic sequences. The decomposition processing module is used to input the aligned chromatographic sequence into the decomposition channel, perform decomposition processing of adaptive baseline fitting and residual signal separation, construct a pure impurity response signal after removing baseline interference, and perform normalization and local enhancement processing to form a cross-batch impurity feature sequence. The anomaly detection module is used to construct a multi-scale difference characterization quantity based on the impurity feature sequence, establish a dynamic control threshold by combining historical batch statistical characteristics, use the dynamic control threshold to perform anomaly signal detection of the multi-scale difference characterization quantity, and output the batch impurity difference identification result.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: By acquiring chromatographic detection data from multiple batches of OLED materials, performing unified sampling processing on the chromatographic detection data, and constructing an original chromatographic sequence with retention time as the sequence index and response intensity as the numerical characterization; based on the original chromatographic sequence, extracting peak shape structural feature parameters from each batch of chromatographic data, the structural feature parameters including peak position, peak width position, and peak shape change trend, and using the peak shape structural feature parameters to construct a corresponding morphology enhancement sequence; after fusing the morphology enhancement sequence and the original chromatographic sequence, constructing a cross-batch reference center spectrum, and performing nonlinear alignment processing of the original chromatographic sequence under the constraint of the cross-batch reference center spectrum to generate an aligned chromatographic sequence; inputting the aligned chromatographic sequence into a decomposition channel, performing adaptive baseline fitting and residual signal separation decomposition processing, constructing a pure impurity response signal after removing baseline interference, and performing normalization and local enhancement processing to form a cross-batch impurity feature sequence; constructing a multi-scale difference characterization quantity based on the impurity feature sequence, and establishing a dynamic control threshold in combination with historical batch statistical characteristics, using the dynamic control threshold to perform abnormal signal judgment of the multi-scale difference characterization quantity, and outputting batch impurity difference identification results. In other words, by extracting peak shape and structural feature parameters from each batch of chromatographic data, the morphology enhancement sequence is fused with the original chromatographic sequence to construct a cross-batch reference center spectrum and perform nonlinear alignment. Adaptive baseline fitting and residual separation are used to extract pure impurity response signals. Combined with multi-scale difference characterization and dynamic control thresholds, automatic anomaly judgment is performed to accurately identify impurity differences down to trace levels, thereby improving the accuracy of OLED material quality control.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application 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 merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the method for identifying impurity differences in multiple batches of OLED materials in this application.

[0020] Figure 2 This is a schematic diagram of the structure of the multi-batch OLED material impurity difference identification system of this application.

[0021] Figure labeling: Unified sampling processing module 11, feature parameter extraction module 12, reference center spectrum construction module 13, decomposition processing module 14, anomaly detection module 15. Detailed Implementation

[0022] This application provides a method and system for identifying impurity differences in multiple batches of OLED materials. This addresses the technical problem in existing technologies where chromatographic retention time drift, baseline interference and noise, and rigid judgment criteria can mask or misjudge true impurity differences between batches, further affecting the accuracy of OLED material quality control. By extracting peak shape and structural characteristic parameters from the chromatographic data of each batch, fusing the morphology enhancement sequence with the original chromatographic sequence, constructing a cross-batch reference center spectrum and performing nonlinear alignment, and employing adaptive baseline fitting and residual separation to extract pure impurity response signals, combined with multi-scale difference characterization quantities and dynamic control thresholds for automatic anomaly detection, the system accurately identifies impurity differences down to trace levels, thus improving the accuracy of OLED material quality control.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0024] Example 1, please refer to the appendix. Figure 1 This application provides a method for identifying impurity differences in multiple batches of OLED materials. The method is applied to a system for identifying impurity differences in multiple batches of OLED materials, and specifically includes the following steps: Acquire chromatographic detection data of multiple batches of OLED materials, perform unified sampling processing on the chromatographic detection data, and construct an original chromatographic sequence with retention time as the sequence index and response intensity as the numerical characterization.

[0025] Specifically, different batches of OLED materials are dissolved in suitable solvents and separated and detected using high-performance liquid chromatography (HPLC). The chromatograph is equipped with a UV-Vis detector or a diode array detector, and fixed mobile phase ratios, column oven temperature, and flow rate are set. After each batch is injected, the detector records the response intensity according to the instrument's own time resolution, such as acquiring 20 data points per second, generating the raw chromatogram for that batch. The raw data for all batches are exported as electronic files, typically including absolute retention times and corresponding response intensities. The chromatographic detection data are the raw signal data recorded by the detector after the OLED materials are separated and detected by HPLC, usually represented as a curve of response intensity changing over time.

[0026] Based on the retention time distribution of the main peaks across all batches, a common retention time analysis interval is determined. For example, if the first impurity peak appears at 1.5 min and the last main peak is completely eluted at 30.0 min, the analysis interval is set to 0.0 min to 30.0 min. A uniform sampling interval is set. The sampling interval is usually the smallest sampling interval in the raw data or an integer fraction thereof. For example, if the original sampling interval of the chromatograph is 0.02 min, the uniform sampling interval is also set to 0.02 min. For each batch of raw data, due to slight differences in the instrument sampling clock, the original retention time points may not be strictly evenly distributed. An interpolation method is used to recalculate the response intensity values ​​at the set uniform sampling time points. For each uniform sampling time point, the two nearest raw data points before and after that time point are found, and the corresponding response intensity at that time point is calculated using a linear interpolation formula. Given the response intensity of the raw data point at the previous time and the response intensity at the next time, as well as the time difference between the two times, the response intensity at any intermediate time is calculated proportionally. After point-by-point interpolation, each batch generates a sequence of the same length. Each position in the sequence corresponds to a fixed retention time point, and the value at each position is the response intensity after interpolation.

[0027] The uniformly sampled results are organized into raw chromatographic sequences with retention time as the row index and response intensity as the numerical value. Multiple sequences are generated for multiple batches, each corresponding to one batch of raw chromatographic data. All sequences have the same length and the same retention time index. Inconsistencies in time points between different batches due to instrument sampling clock drift or differences in data export formats are eliminated, and chromatographic data from all batches are compared at the same time points.

[0028] Based on the original chromatographic sequence, peak shape structural feature parameters are extracted from each batch of chromatographic data. The peak shape structural feature parameters include peak position, peak width position, and peak shape change trend. Corresponding morphology enhancement sequences are constructed using the peak shape structural feature parameters.

[0029] Furthermore, this application also includes the following steps: calculating the first-order gradient sequence and the second-order curvature sequence along the retention time direction of the original chromatographic sequence, and dividing the chromatographic signal into candidate intervals according to the position of gradient sign change to determine the time boundary of each peak structure; within each peak candidate interval, determining the peak position based on the maximum response pad of the peak region, and determining the peak width position based on the time span corresponding to the signal attenuation to a preset proportional threshold on both sides of the peak position, and constructing peak shape asymmetry parameters according to the gradient integral difference on both sides of the peak position; coupling and fusing the peak position, peak width position, peak shape asymmetry parameters and curvature change trend parameters, and using the coupling and fusion results to perform local weighted reconstruction of the original chromatographic sequence to obtain a morphology-enhanced sequence.

[0030] Specifically, for the constructed original chromatographic sequence, starting from the second time point to the last time point, the difference in response intensity between the current point and the previous point is calculated sequentially, and then divided by the sampling interval to obtain the first-order gradient value for that point. The gradient value for the first point in the sequence is set to zero. After calculation, a first-order gradient sequence of the same length as the original chromatographic sequence is obtained. Positive values ​​represent the rising phase of the chromatographic signal, negative values ​​represent the falling phase, and values ​​close to 0 represent peaks or troughs. The first-order gradient sequence is a sequence formed by calculating the rate of change of the response intensity value at each time point in the original chromatographic sequence along the retention time direction, describing the instantaneous rate of change of the chromatographic signal at each time point.

[0031] The same method is used to differentiate the first-order gradient sequence again. Starting from the second gradient point, the difference between the current gradient value and the previous gradient value is calculated and divided by the sampling interval to obtain the second-order curvature value. The curvature of the first point is set to zero. The second-order curvature sequence reflects the degree of signal bending. In the rising section of the chromatographic peak, the first-order gradient is positive and gradually increases, and the second-order curvature is positive, indicating that the curve bends upward, i.e., it is concave. Near the peak, the first-order gradient changes from positive to negative, and the second-order curvature changes from positive to negative, with the second-order curvature being negative and having the largest absolute value at the peak. In the falling section, the first-order gradient is negative and its absolute value gradually decreases, and the second-order curvature is positive. In the flat baseline section, the second-order curvature is close to 0.

[0032] Traverse the first-order gradient sequence, recording the points where the gradient value changes from negative to positive as peak start points; record the points where the gradient value changes from positive to negative as peak apexes; record the points where the gradient value changes from positive to negative and then back to positive as peak end points. Start the search from the beginning of the first-order gradient sequence. When a gradient value changes from less than 0 to greater than 0, mark that position as a peak start point; continue the search, marking the point where the gradient value changes from greater than 0 to less than 0 as a peak apex; continue the search, marking the point where the gradient value changes from less than 0 to greater than 0 again as a peak end point. A complete candidate interval consists of the peak start point to the peak end point, containing a peak apex in between. If multiple closely connected peaks appear, each peak is divided into intervals according to the above rules. Isolated small noise fluctuations, due to irregular gradient sign changes or failure to meet peak shape characteristics, can be eliminated later.

[0033] For each candidate interval, find the time point with the strongest response intensity between the start and end points of the interval; the corresponding retention time is the peak position. If multiple points have equal and maximum response intensities, take the median of these points as the peak position.

[0034] Calculate the peak height, which is the response intensity at the peak position minus the baseline response intensity within that interval. The baseline intensity can be the average of the response intensities at the start and end points of the interval. Set a proportional threshold, such as 50%. Calculate the response intensity value corresponding to half the peak height, which is the baseline intensity plus half the peak height. Search to the left from the peak position to find the first time point where the response intensity is less than or equal to the half-peak intensity; this point is recorded as the left half-peak width point. Similarly, search to the right to find the first time point where the response intensity is less than or equal to the half-peak intensity; this is recorded as the right half-peak width point. The time difference between the left and right half-peak width points is the peak width. The peak width position is the time span on both sides of the peak corresponding to half the peak height or a specified proportion of the peak height. Usually, with the peak position as the center, find the time points on both sides where the signal decays to a certain proportion of the peak height, such as 50%; the time difference between these two points is the peak width.

[0035] Within the interval to the left of the peak, the first-order gradient value is integrated, i.e., the gradient values ​​at each point are multiplied by the sampling interval, to obtain the cumulative change in the rising segment on the left. Similarly, within the interval to the right of the peak, the absolute value of the first-order gradient is integrated to obtain the cumulative change in the falling segment on the right. The ratio of the left-side integral to the right-side integral is calculated. If the ratio is close to 1, it indicates symmetry between the left and right sides; if the ratio is greater than 1, it indicates a gentler rise on the left and a steeper fall on the right, indicating a leading peak; if the ratio is less than 1, it indicates a steeper rise on the left and a gentler fall on the right, indicating a tailing peak. This ratio is used as the peak shape asymmetry parameter. The peak shape asymmetry parameter is a quantitative indicator describing whether the shapes of the left and right sides of a chromatographic peak are symmetrical. By comparing the magnitudes of the gradient integrals on the left and right sides of the peak, it can be determined whether the peak is leading (gentle on the left, steep on the right) or tailing (steep on the left, gentle on the right).

[0036] Within the same candidate interval, observe the changing trend of the second-order curvature sequence. Record the time points when the curvature changes from positive to negative, corresponding to the inflection point of the transition from the peak's rising segment to the peak, and the time points when the curvature changes from negative to positive, corresponding to the inflection point of the transition from the peak to the falling segment. Simultaneously, calculate the minimum curvature near the peak and the offset of the time position of this minimum relative to the peak position; these together constitute the curvature change trend parameter, used to describe the sharpness and bending characteristics of the peak shape.

[0037] Peak position, peak width, peak shape asymmetry parameters, and curvature trend parameters are coupled and fused. For each time point in the original chromatographic sequence, a local enhancement weighting coefficient is calculated based on its relationship with the peak position and width within the candidate interval, as well as asymmetry and curvature information. The weighting coefficient is highest near the peak position; lower within the peak width range; and further decreases outside the peak width range but within the candidate interval. On the side with stronger asymmetry, the weighting coefficient is appropriately increased to compensate for the peak shape asymmetry. Simultaneously, based on the curvature trend, regions with large negative curvature values ​​are given additional high weights to highlight the precise position of the peak apex. The original response intensity at each time point is multiplied by the corresponding local enhancement weighting coefficient to obtain the enhanced response intensity. After point-by-point processing, the enhanced response intensities at all time points constitute a new sequence, namely the morphological enhancement sequence. The morphological enhancement sequence maintains the same retention time index as the original chromatographic sequence, and the value at each time point is the result of weighted enhancement of the original intensity. In the morphological enhancement sequence, the true chromatographic peak is effectively amplified, while baseline noise and minor perturbations are relatively suppressed. By using peak position, peak width, asymmetry, and curvature parameters for local weighting, the response intensity of the true chromatographic peak is enhanced, while baseline fluctuations and random noise are relatively suppressed due to their lower weights, thus improving the signal-to-noise ratio.

[0038] After fusing the morphological enhancement sequence and the original chromatographic sequence, a cross-batch reference center spectrum is constructed, and nonlinear alignment processing of the original chromatographic sequence is performed under the constraint of the cross-batch reference center spectrum to generate an aligned chromatographic sequence.

[0039] Furthermore, this application also includes the following steps: weighted fusion of the morphological enhancement sequences of each batch with the corresponding original chromatographic sequences to form a fusion sequence that highlights the peak shape and structural features; extraction of the corresponding peak positions and peak structure intervals of each batch based on the fusion sequence, and construction of a cross-batch peak structure correspondence set by matching and aligning the same peak structures in different batches; under the constraints of the peak structure correspondence set, weighted superposition and position correction processing are performed on the fusion sequences of each batch within the corresponding peak structure intervals, and smooth fusion is performed on non-peak structure intervals to generate a cross-batch reference center spectrum with uniform peak structure distribution features.

[0040] Specifically, each batch contains two sequences: an unprocessed raw chromatographic sequence and a morphology-enhanced sequence with peak shape enhancement. The raw chromatographic and morphology-enhanced sequences from the same batch are combined in a specific ratio to generate a new fusion sequence. The fusion sequence retains the integrity of the original signal while highlighting the peak shape characteristics. For each retention time point, the original response intensity is multiplied by the original weighting coefficient, and then the morphology-enhanced response intensity is multiplied by the enhancement weighting coefficient. The sum is then divided by the sum of the two weighting coefficients. The original weighting coefficient is typically set to 0.5, and the enhancement weighting coefficient is also set to 0.5, i.e., the arithmetic mean of the two is taken. If the original signal noise is high in certain regions, the enhancement weighting coefficient is appropriately increased, such as setting it to 0.7, and the original weighting coefficient to 0.3. After point-by-point weighted calculation, a new fusion sequence is generated for each batch.

[0041] For each batch of fusion sequences, the system automatically identifies the peak positions of all chromatographic peaks in that batch, as well as the start and end times of each peak. Each peak structure interval is represented by three time values: start time, peak position time, and end time. The peak structure interval is the retention time range of a chromatographic peak from its start point to its end point, including the rising, peak apex, and falling portion of the peak. Due to the drift in retention times between different batches, matching cannot be done directly based on retention time values; instead, multiple features must be considered for matching. All peaks from all batches are roughly sorted according to their peak position time. A matching window is set, and for each peak in the first batch, all peaks in the second batch whose peak position times fall within the matching window range are searched. If multiple candidate peaks are found, the similarity of peak width and peak shape asymmetry parameters are further compared. Peaks that meet all conditions are considered to correspond to the same substance. Matching peaks from all batches are grouped into the same correspondence group. A correspondence group contains multiple peaks, each from a different batch, all representing the same component. For peaks that may be missing in certain batches, i.e., those impurities not detected in the batch, they are marked as missing in the corresponding relationship group. All relationship groups constitute a cross-batch peak structure relationship set. The peak structure relationship set is a collection of information recording the peak structures corresponding to the same chemical component in chromatographic data from all batches.

[0042] For each peak structure correspondence group, fused sequence fragments of that peak from all batches are collected. Due to retention time drift, the peak positions of these fragments are not identical. The median of all peak positions within each correspondence group is calculated as the reference peak position. For each batch of peak fragments, a linear interpolation method is used to shift or scale the time axis of the entire fragment so that the peak position of the fragment is precisely moved to the reference peak position, while maintaining the relative proportion of peak width. After correction, the peak fragments of all batches are aligned on the time axis. The aligned peak fragments are then weighted and superimposed, with each batch having an equal weight, meaning each batch contributes equally. The response intensity of each batch at the corrected time point is multiplied by its respective weight, summed, and then divided by the total weight to obtain the average response intensity at the reference peak position. This operation is repeated for all time points to obtain the reference peak shape representing that component.

[0043] For baseline regions without peak structure, i.e., retention time ranges that do not belong to any peak structure intervals, complex stacking and correction are not required. The baseline reference value is obtained by simply averaging the response intensities of all batches of fusion sequences within these regions. To maintain the continuity of the reference center spectrum, a smooth transition is implemented at the boundary between peak and baseline intervals, such as using a linear gradient, to ensure the overall reference center spectrum is smooth and without abrupt changes.

[0044] Within the corresponding peak structure range, the peak positions of each batch are adjusted to a uniform time position, resulting in a more concentrated peak shape after stacking. For baseline regions without peak structure, complex stacking and correction are not performed; instead, simple smoothing is used to ensure the continuity and smoothness of the reference central spectrum in non-peak regions. The various reference peak shapes and smoothed baselines are then spliced ​​together in retention time order to form a complete chromatographic curve, i.e., the cross-batch reference central spectrum.

[0045] For each batch of raw chromatographic sequences, nonlinear alignment with cross-batch reference central spectra is required. A time transformation function is sought that ensures the peak positions of the raw sequences are as consistent as possible with those of the reference central spectra after transformation. The retention times of all peaks in each batch's raw sequences and the corresponding peaks in the reference central spectra are extracted, representing the fixed reference peak positions. These two sets of time points are used as control point pairs. A smooth time transformation function is constructed using piecewise polynomial interpolation, mapping each retention time of the raw sequences to a new retention time, thus aligning the raw peak positions with the reference peak positions. For regions without peak control points, the transformation function remains smooth and as close to linear as possible. The time transformation function is applied to each retention time point of the raw chromatographic sequences, redistributing the raw response intensity to the new retention time positions. Simultaneously, interpolation ensures evenly spaced sampling of the new sequences, resulting in the nonlinearly aligned chromatographic sequences for that batch. The aforementioned steps are repeated for all batches to obtain the aligned chromatographic sequences for all batches. In the aligned chromatographic sequence, the peaks of the same substance appear at the same or very close retention time positions as the reference central spectrum, thus achieving precise comparability between batches.

[0046] The aligned chromatographic sequence is input into the decomposition channel, and decomposition processing with adaptive baseline fitting and residual signal separation is performed to construct a pure impurity response signal after removing baseline interference. Normalization and local enhancement processing are then performed to form a cross-batch impurity characteristic sequence.

[0047] Furthermore, this application also includes the following steps: constructing a sliding analysis window along the retention time direction for the aligned chromatographic sequence; extracting a set of local minimum points of characteristic signal intensity within each sliding analysis window; generating an initial baseline control node sequence based on the set of local minimum points; constructing an initial baseline estimation curve by piecewise interpolation fitting of the initial baseline control node sequence; performing baseline stripping processing on the aligned chromatographic sequence using the initial baseline estimation curve as a constrained envelope to obtain residual signals; and calculating the rate of change of signal gradient and rate of change of curvature based on the continuity of the residual signals between adjacent sliding analysis windows. A continuity discrimination index characterizing the stability of peak-shaped structures is constructed. Based on the continuity discrimination index, structural components are screened from the residual signal. Signal segments that simultaneously satisfy the amplitude continuity constraint, gradient monotonic change constraint, and curvature consistency constraint are identified as structural response components, while the remaining discrete high-frequency disturbance signals are eliminated as random noise components. Cross-batch scale uniform processing is performed on the structural response components. The amplitude of each batch of structural response components is normalized by the peak intensity ratio mapping relationship based on the cross-batch reference center spectrum. Signal contrast enhancement is performed in the local time neighborhood corresponding to each structural response component to form an impurity feature sequence.

[0048] Furthermore, this application also includes the following steps: within each sliding analysis window, a first-order gradient sequence and a second-order curvature sequence are calculated based on the residual signal, and a continuity discrimination index is constructed in combination with the signal amplitude change. The continuity discrimination index is used to characterize the continuity of the signal amplitude, the consistency of the gradient change direction, and the stability of the curvature change in the local time neighborhood.

[0049] Furthermore, this application also includes the following steps: using the continuity discrimination index as a constraint, performing candidate structure segment division on the residual signal, identifying signal segments whose continuity discrimination index remains stable within a preset interval as candidate structure segments; performing amplitude continuity constraint verification on the candidate structure segments to establish a first verification result; performing monotonicity interval division on the gradient sequence of the candidate structure segments to establish a second verification result; verifying the curvature consistency of the peak region of the candidate structure segments based on the curvature sequence change trend to generate a third verification result; when the first verification result, the second verification result, and the third verification result are all passing results, then a structural response component is established.

[0050] Furthermore, this application also includes the following steps: performing cross-window consistency verification on the structural response components within adjacent sliding analysis windows, identifying stable and drifting segments of the structural response components by comparing peak position offset, peak width change rate, and gradient direction consistency in adjacent sliding analysis windows; performing local time axis correction processing based on the stability decay degree of the drifting segments; and performing cross-batch scale unification processing based on the correction processing results and the stable segments.

[0051] Specifically, the width of the sliding window and the sliding step size are determined based on the average width of the chromatographic peak and the analytical precision. The window width is typically set to two to three times the full width at half maximum (FWHM) of a typical chromatographic peak, such as 0.2 min. The sliding step size is typically set to an integer multiple of the sampling interval, such as 0.01 min, to ensure a large overlap between adjacent windows, with an overlap rate of up to 95%, guaranteeing a smooth transition in baseline fitting. Starting from the beginning of the chromatographic sequence, data segments within each window are extracted sequentially. The first window covers from 0 to 0.2 min, and the window is then slid to the right by 0.01 min. The second window covers from 0.1 min to 0.21 min, and so on, until the right end of the window covers the end of the sequence. For each window, all retention time points within that window and their corresponding response intensities are extracted.

[0052] For each sliding analysis window, a local minimum is identified by iterating through all internal data points where the response intensity is simultaneously less than that of its left and right adjacent points. All points meeting this condition are recorded, forming the set of local minimum points for that window. Since peak apex represents a maxima, and small fluctuations in valleys and baseline flat regions can also produce minima, multiple minimum points can typically be extracted for each window.

[0053] All local minima extracted by the sliding windows are merged together and sorted by retention time. Due to window overlap, the same minimum point may be extracted repeatedly by multiple windows, so deduplication is required; only one of the identical or extremely close points is retained. Outliers with response intensities significantly higher than the surrounding baseline region are removed. The resulting point sequence is the initial baseline control node sequence, representing the lowest point in the chromatographic signal that may belong to the baseline.

[0054] The entire retention time axis is divided into several small intervals, with the endpoints of each interval determined by two adjacent control nodes. Within each interval, a smooth curve passing through both endpoints and exhibiting continuous first and second derivatives is fitted using cubic spline interpolation. The polynomial coefficients are solved using the conditions of equal function values ​​at the nodes and continuous first and second derivatives. Connecting the cubic polynomials of all intervals yields an overall smooth curve, which is the initial baseline estimation curve, located below the chromatographic signal and reflecting the baseline drift trend over time.

[0055] For each retention time point, the value of the initial baseline estimate curve at that point is subtracted from the original response intensity of the aligned chromatographic sequence. If the difference is less than 0, the residual signal at that point is forced to 0. If the difference is greater than 0, the difference is retained. After point-by-point calculation, a new sequence is obtained, called the residual signal. In the residual signal, the original baseline drift has been subtracted, leaving mainly the chromatographic peak signal and the random noise superimposed on it.

[0056] For two adjacent windows, their overlapping region is selected. Within this overlapping region, the first-order gradient sequence and second-order curvature sequence of the residual signal for one window are calculated, and the corresponding sequences for the other window are calculated similarly. The absolute value of the gradient difference between the two windows at the same time point is calculated and divided by the sampling interval to obtain the gradient rate of change; similarly, the absolute value of the curvature difference is calculated and divided by the sampling interval to obtain the curvature rate of change. The average of the rates of change over all overlapping time points is taken to obtain the gradient rate of change and curvature rate of change for the adjacent window pair.

[0057] For each residual signal segment within a sliding window, the first-order gradient and second-order curvature are calculated. For each internal point within the window, the first-order gradient value is obtained by subtracting the intensity of the previous point from the intensity of the next point and dividing by the sampling interval. The second-order curvature value is obtained by differentiating the first-order gradient sequence again.

[0058] Within a sliding window, for each candidate time point, a continuity index characterizing the stability of the peak structure is constructed. The continuity of amplitude is quantified by examining the smoothness of response intensity changes at each point and in its neighborhood, calculating the variance or maximum / minimum difference of the response intensity within the neighborhood. A smaller variance indicates more continuous amplitude. The consistency of gradient direction is quantified by examining whether the sign of the first-order gradient near the point remains stable. For the rising segment of a chromatographic peak, the gradient should be positive and change gradually; the falling segment should have a negative gradient and change gradually; the gradient near the peak should change from positive to negative, and the sign change point should be unique. The number of gradient sign changes within the neighborhood is calculated. Frequent alternation between positive and negative signs indicates severe noise interference. The consistency of gradient direction is quantified into a single value, with a higher value for consistent signs and a lower value for frequent changes. The stability of curvature changes is quantified by judging the fluctuation of the second-order curvature near the point. In the rising segment of the chromatographic peak, the curvature should be positive and gradually decrease; the curvature near the peak should change from positive to negative; the curvature in the falling segment should be either positive or negative. Calculate the standard deviation or range of the curvature values ​​within the neighborhood. The smaller the curvature variation, the more stable the peak shape.

[0059] The continuity of amplitude, consistency of gradient direction, and stability of curvature are weighted and summed according to certain weights, such as each accounting for 1 / 3, to obtain the continuity discrimination index for that point. This index is used to determine whether a signal segment belongs to a stable chromatographic peak structure. The closer the continuity discrimination index value is to one, the more the signal segment conforms to the structural characteristics of an ideal chromatographic peak; the closer it is to zero, the more likely the segment is random noise or baseline fluctuation. For example, to quantify the continuity of amplitude, the intensities of five points from 5.23 to 5.27 are taken as 12.70, 16.83, 21.36, 22.38, and 20.64. The calculated mean is 18.78, the variance is 12.77, and the maximum and minimum difference is 9.68. After normalization, the amplitude continuity score is 0.92. To quantify the consistency of gradient direction, the gradients from 5.23 to 5.27 are checked as 505, 413, 453, 102, and -174. The sign change is positive → positive → positive → positive → negative, changing only once (from positive to negative), and it is located near the peak, which is reasonable. The gradient direction consistency score is 1.0. Quantifying the stability of curvature changes, the curvature values ​​are 25200, -9200, 4000, -35100, and -27600, showing significant variations, but the curvature changing from positive to negative near the peak is normal. The calculated standard deviation of curvature in the neighborhood is approximately 22000, and the normalized score is 0.65. With a comprehensive weight of 1 / 3, the continuity discriminant index is 0.857. The high continuity discriminant index value indicates that this point belongs to a stable chromatographic peak structure. For the baseline noise region (e.g., 5.00 minutes), the residual intensity is close to 0, the gradient alternates randomly between positive and negative, the curvature fluctuates drastically, and the continuity discriminant index is below 0.2.

[0060] Based on historical experience, a preset threshold interval is determined to distinguish stable structural signals from random noise. For each retention time point in the residual signal, the corresponding continuity discrimination index value is obtained. If it falls within the preset threshold interval, the point is marked as a candidate point. Consecutive adjacent candidate points are merged into segments. If all index values ​​in a segment are within the preset threshold interval and the segment length is greater than a certain value, the segment is identified as a candidate structural segment. Regions where index values ​​are not within the preset threshold interval are temporarily set aside.

[0061] For each candidate structural segment, check the continuity of its internal response intensity. Calculate the absolute value of the intensity difference between all adjacent sampling points within the segment and find the maximum value. Calculate the overall standard deviation of the intensity values ​​within the segment. If the maximum adjacent difference is less than a preset jump threshold, and the ratio of the standard deviation to the average intensity of the segment is less than a preset ratio, then the segment is considered to meet the amplitude continuity constraint, and the first verification result is passed; otherwise, it is failed.

[0062] For candidate fragments that pass the first check, the first-order gradient sequence of the residual signal within the fragment is extracted. Based on the sign of the gradient, the fragment is divided into several monotonic intervals. Starting from the fragment's origin, regions with continuous positive gradients are designated as ascending intervals, regions with continuous negative gradients as descending intervals, and regions with gradients close to 0 as plateau intervals. A typical chromatographic peak should contain one ascending interval and one descending interval, with the ascending interval preceding the descending interval, and a short plateau interval in between. First, both ascending and descending intervals must exist, and the total number of intervals cannot exceed three. Second, within each monotonic interval, the gradient value cannot exhibit sign reversal, i.e., it cannot alternate between positive and negative. Third, the gradient change within the ascending interval should first increase and then decrease, and the absolute value of the gradient within the descending interval should first increase and then decrease. When all three conditions are met, the second check result is considered passed; otherwise, it is considered failed.

[0063] For candidate segments that pass the second check, extract the second-order curvature sequence within the segment. The check process is as follows: First, to the left of the peak, the curvature value should be positive; to the right of the peak, the curvature value may be negative or positive, but there should be a process of changing from positive to negative and then back to positive; Second, near the peak, the curvature value should reach a negative minimum, i.e., the absolute value should be maximum; Third, the change in the curvature sequence should be smooth, without any violent oscillations. Find the point where the gradient changes from positive to negative, and check the curvature values ​​within three points before and after that point. The curvature before the peak should be positive, and the curvature after the peak should be negative, with the curvature at the peak being negative and having a large absolute value. If the above trend is met, the third check result is passed; otherwise, it is failed.

[0064] For a candidate structural fragment, it is only formally recognized as a structural response component if the first, second, and third verification results all pass. A structural response component represents a genuine chromatographic peak in the residual signal within the fragment, rather than baseline fluctuations or random noise. All fragments that pass the triple verification are retained and labeled as structural response components. All residual signal portions not labeled as structural response components, including candidate fragments that fail any verification, regions with continuity discrimination indices below the threshold, and regions with excessively short fragment lengths, are classified as random noise components and do not contain useful impurity information. A label is generated for each retention time point; if it belongs to a structural response component, it is labeled as a structure; otherwise, it is labeled as noise.

[0065] Based on the actual data accuracy requirements and allowable fluctuation range, allowable thresholds for peak position offset, peak width change rate, and gradient direction consistency are set. For each pair of adjacent windows, it is determined whether the two windows contain the same structural response component, i.e., the same impurity peak. This is determined by the difference between the peak position of this component in the preceding window and the peak position in the following window of each pair of adjacent windows being less than a larger window value, such as 0.15 min, and the peak widths being similar. If it is confirmed to be the same component, the peak position time and peak width of the two windows are extracted, and the peak position offset and peak width change rate are calculated. The peak position offset is obtained by calculating the difference in peak position time between the two windows; the peak width change rate is obtained by dividing the difference in peak width between the two windows by the peak width of the preceding window in each pair of adjacent windows. Within the overlapping region, the proportion of points with the same gradient sign is counted. If it is ≥90%, the gradient direction is considered consistent. If the peak offset is ≤0.02 min, the peak width change rate is ≤10%, and the proportion of points with the same gradient sign is ≥90% of the total number of points, then the structural response component is considered stable between adjacent window pairs. The time regions covered by these window pairs are marked as stable segments. If any indicator exceeds the threshold, drift is considered to exist, and the time regions covered by these window pairs are marked as drift segments.

[0066] The degree of stability degradation is quantified by the number of times the peak position shift exceeds a threshold or the number of times the peak width change rate exceeds a threshold. The degradation degree of the peak width change rate is calculated as (actual change rate - 10%) / 10%. The overall degradation degree can be the maximum of these two values ​​or a weighted sum. This value will be used to determine the intensity of the local time-axis correction. The peak positions within the drift segment are pulled back to positions consistent with the reference center spectrum or adjacent stable segments. The boundaries of the drift segment are determined, and the stiffness of the correction is determined based on the degree of degradation. The greater the degradation degree, the greater the correction magnitude. The mapping function is applied to each retention time point within the drift segment to obtain the corrected retention time. For stable segments, no time-axis correction is performed; the original retention time is directly retained. After completing the local time-axis correction, the structural response components of all batches are highly consistent on the time axis.

[0067] The identified structural response components undergo cross-batch scale unification processing. Based on the cross-batch reference center spectrum, the peak area of ​​each impurity peak in the cross-batch reference center spectrum is extracted as a baseline value. For the corresponding structural response component in each batch, the ratio of its peak area to the corresponding peak area in the reference center spectrum is calculated to obtain a scaling factor. All intensity values ​​of the entire structural response component in that batch are multiplied by the reciprocal of this scaling factor to ensure that the amplitude of the impurity peaks in that batch is consistent with the reference center spectrum. Signal contrast enhancement is performed within the local time neighborhood corresponding to each structural response component. First, the minimum and maximum intensity values ​​within the local neighborhood are calculated. Then, the intensity value of each point is enhanced according to the formula: Enhanced Intensity = (Original Intensity - Minimum) / (Maximum - Minimum) × Output Range + Offset. The output range is between 0 and 100, and the offset is a constant added to or subtracted from the normalized and scaled intensity values ​​in the contrast enhancement formula. It does not change the relative contrast of the signal but can adjust the baseline position of the enhanced signal or make it fall into a specific numerical range, usually 0 or 5. After enhancement, the contrast between the impurity peaks and the surrounding background is significantly improved, and the weak impurity signal becomes more obvious.

[0068] The normalized and contrast-enhanced structural response components are reorganized into a complete sequence according to retention time. For regions of removed random noise components, their intensity values ​​are set to zero or a very small constant, resulting in the impurity characteristic sequence. A sliding window local minimum point set combined with cubic spline interpolation can adaptively fit complex and varying baselines, accurately subtracting even nonlinear drift in the baseline, ensuring that only the true impurity response is retained in the residual signal. Amplitude normalization based on peak intensity scaling of the reference center spectrum eliminates batch-to-batch amplitude differences caused by fluctuations in experimental conditions such as injection volume and detector sensitivity, giving impurity characteristic sequences from different batches a uniform scale for direct comparison.

[0069] A multi-scale difference characterization quantity is constructed based on the impurity feature sequence, and a dynamic control threshold is established by combining historical batch statistical features. The abnormal signal judgment of the multi-scale difference characterization quantity is performed using the dynamic control threshold, and the batch impurity difference identification result is output.

[0070] Furthermore, this application also includes the following steps: reading the multi-scale difference characterization quantity in the statistical characteristics of historical batches, constructing a time series distribution model using the multi-scale difference characterization quantity; evaluating the degree of statistical distribution drift between different time batches using the time series distribution model, performing dynamic adaptive correction of the control threshold based on the evaluation result, and establishing a dynamic control threshold.

[0071] Furthermore, this application also includes the following steps: configuring a batch anomaly identifier based on the batch impurity difference identification result, and using the batch anomaly identifier to perform anomaly diversion management of multiple batches of OLED materials.

[0072] Specifically, for each batch, multiple scales of differential characterization measures are extracted from the impurity feature sequence, including but not limited to the peak height, peak area, peak width, and asymmetry factor of a single impurity peak; the relative proportions between multiple impurity peaks; the change in the number of impurity peaks; and the overall similarity measure across the entire spectrum.

[0073] The process involves acquiring historical batch statistical characteristics, specifically calculating the impurity characteristic sequences of multiple batches of materials that have been confirmed as qualified over a past period. This yields statistical measures such as the mean, variance, distribution type, and percentiles of each variance characteristic. For each variance characteristic, a time series is constructed with the batch production time as the x-axis and the value of the characteristic as the y-axis. A sliding window method is used to calculate local statistical characteristics near each time point, including the local mean, local standard deviation, and local distribution pattern. These statistical characteristics are combined into a time series distribution model to describe the normal fluctuation range and trend of the variance characteristics over time. This model can be expressed as follows: at time point t, the predicted mean of a characteristic is μ(t), and the predicted standard deviation is σ(t), where σ(t) can change slowly over time. For each time point, there are parameters describing its recent normal distribution, namely the mean and standard deviation.

[0074] For the current batch to be tested, extract its multi-scale difference characterization vector, compare this vector with the time series distribution model, and calculate the degree of statistical distribution drift. Mean drift = |current batch characterization - local mean at the current time point| / local standard deviation. Based on the drift assessment results, perform dynamic adaptive correction of the control threshold. If the drift is small, it indicates process stability, and the control threshold can be kept relatively tight to improve sensitivity to minor anomalies; if the drift is moderate, it indicates that the process may have a slow, acceptable trend change, and the control threshold can be moderately relaxed to avoid frequent false alarms caused by normal process drift; if the drift is large, it indicates that the process may have an abnormal change, and the threshold should not be relaxed but rather tightened and an early warning should be triggered to prompt operators to check the process. For each difference characterization, establish its own dynamic control threshold. Since different characterizations have different fluctuation characteristics and different degrees of influence on material properties, their control thresholds can be set independently.

[0075] Each multi-scale difference characterization quantity of the current batch is compared with its corresponding dynamic control threshold. If the value of a multi-scale difference characterization quantity is greater than the upper control limit or less than the lower limit of the dynamic control threshold, the characterization quantity is determined to be out of limit; if all characterization quantities are within limits, the batch is determined to be normal; if one or more characterization quantities are out of limit, the batch is determined to be abnormal. For batches determined to be abnormal, further analysis is conducted to determine which characterization quantities caused the exceedance and the extent of the exceedance, forming a batch impurity difference identification result, including batch number, judgment conclusion (normal / abnormal), list of characterization quantities exceeding limits and their values, the extent of the exceedance (e.g., percentage exceeding the upper control limit), and recommended follow-up measures.

[0076] Based on the batch impurity difference identification results, each batch is assigned a batch anomaly label, such as qualified, pending re-inspection, unqualified, or requiring traceability. The label automatically triggers the corresponding material diversion operation. Qualified batches automatically generate release slips, allowing the materials to enter the next production stage; batches pending re-inspection automatically create re-inspection tasks, notifying quality inspectors to resample and test, while the batch of materials is temporarily stored in the inspection area and not allowed to enter the next stage; unqualified batches are automatically locked, prohibiting their release, and a non-conforming product report is generated, notifying the quality engineer and production department to investigate the cause; batches requiring traceability send warning information to the process engineer, reminding them to pay attention to potential drift in process parameters. The impurity difference identification results, anomaly labels, and diversion processing records for each batch are stored in a database as a data source for subsequent updates to historical batch statistical characteristics, forming a closed-loop quality control system.

[0077] Extracting differential characterization parameters at three scales—single-peak, multi-peak, and full-spectrum—enables the capture of variations in different types of impurities, improving the comprehensiveness and sensitivity of anomaly detection. Linking the identification results with the production management system automatically configures anomaly flags and triggers corresponding material diversion operations, achieving closed-loop control from detection to decision-making, significantly improving quality management efficiency and response speed.

[0078] In summary, the multi-batch OLED material impurity difference identification method provided in this application has the following technical effects: By acquiring chromatographic detection data of multiple batches of OLED materials, performing unified sampling processing of the chromatographic detection data, and constructing an original chromatographic sequence with retention time as the sequence index and response intensity as the numerical characterization; based on the original chromatographic sequence, extracting peak shape structural feature parameters from the chromatographic data of each batch, including peak position, peak width position, and peak shape change trend, and using the peak shape structural feature parameters to construct a corresponding morphology enhancement sequence; after fusing the morphology enhancement sequence and the original chromatographic sequence, constructing a cross-batch reference center spectrum, and performing nonlinear alignment processing of the original chromatographic sequence under the constraint of the cross-batch reference center spectrum to generate an aligned chromatographic sequence; inputting the aligned chromatographic sequence into the decomposition channel, performing adaptive baseline fitting and residual signal separation decomposition processing, constructing a pure impurity response signal after removing baseline interference, and performing normalization and local enhancement processing to form a cross-batch impurity feature sequence; constructing a multi-scale difference characterization quantity based on the impurity feature sequence, and establishing a dynamic control threshold in combination with historical batch statistical characteristics, using the dynamic control threshold to perform abnormal signal judgment of the multi-scale difference characterization quantity, and outputting the batch impurity difference identification result. In other words, by extracting peak shape and structural feature parameters from each batch of chromatographic data, the morphology enhancement sequence is fused with the original chromatographic sequence to construct a cross-batch reference center spectrum and perform nonlinear alignment. Adaptive baseline fitting and residual separation are used to extract pure impurity response signals. Combined with multi-scale difference characterization and dynamic control thresholds, automatic anomaly judgment is performed to accurately identify impurity differences down to trace levels, thereby improving the accuracy of OLED material quality control.

[0079] Example 2: Based on the same inventive concept as the multi-batch OLED material impurity difference identification method in Example 1, this application also provides a multi-batch OLED material impurity difference identification system. Please refer to the appendix. Figure 2The multi-batch OLED material impurity difference identification system includes: a unified sampling processing module 11, used to acquire chromatographic detection data of multiple batches of OLED materials, perform unified sampling processing of the chromatographic detection data, and construct an original chromatographic sequence with retention time as the sequence index and response intensity as the numerical characterization; a feature parameter extraction module 12, used to extract peak shape structure feature parameters from each batch of chromatographic data based on the original chromatographic sequence, the structure feature parameters including peak position, peak width position, and peak shape change trend, and construct a corresponding morphology enhancement sequence using the peak shape structure feature parameters; and a reference center spectrum construction module 13, used to fuse the morphology enhancement sequence and the original chromatographic sequence... A cross-batch reference center spectrum is constructed, and nonlinear alignment processing of the original chromatographic sequence is performed under the constraint of the cross-batch reference center spectrum to generate aligned chromatographic sequences. The decomposition processing module 14 is used to input the aligned chromatographic sequences into the decomposition channel, perform decomposition processing of adaptive baseline fitting and residual signal separation, construct a pure impurity response signal after removing baseline interference, and perform normalization and local enhancement processing to form a cross-batch impurity feature sequence. The anomaly judgment module 15 is used to construct a multi-scale difference characterization quantity based on the impurity feature sequence, and establish a dynamic control threshold by combining historical batch statistical characteristics. The dynamic control threshold is used to perform anomaly signal judgment of the multi-scale difference characterization quantity, and output the batch impurity difference identification result.

[0080] Furthermore, the feature parameter extraction module 12 in the multi-batch OLED material impurity difference identification system is also used for: calculating the first-order gradient sequence and the second-order curvature sequence along the retention time direction of the original chromatographic sequence, and dividing the chromatographic signal into candidate intervals according to the gradient sign change position to determine the time boundary of each peak structure; within each peak candidate interval, determining the peak position based on the maximum response pad of the peak region, and determining the peak width position based on the time span corresponding to the signal attenuation to a preset ratio threshold on both sides of the peak position, and constructing peak shape asymmetry parameters according to the gradient integral difference on both sides of the peak position; coupling and fusing the peak position, peak width position, peak shape asymmetry parameters and curvature change trend parameters, and using the coupling and fusion results to perform local weighted reconstruction of the original chromatographic sequence to obtain the morphology-enhanced sequence.

[0081] Furthermore, the reference center spectrum construction module 13 in the multi-batch OLED material impurity difference identification system is also used to: weightedly fuse the morphology enhancement sequences of each batch with the corresponding original chromatographic sequences to form a fusion sequence that highlights the peak shape structure features; extract the corresponding peak positions and peak structure intervals of each batch based on the fusion sequence, and construct a cross-batch peak structure correspondence set by matching and aligning the same peak structures in different batches; under the constraints of the peak structure correspondence set, perform weighted superposition and position correction processing on the fusion sequences of each batch in the corresponding peak structure intervals, and perform smooth fusion on the non-peak structure intervals to generate a cross-batch reference center spectrum with uniform peak structure distribution features.

[0082] Furthermore, the decomposition processing module 14 in the multi-batch OLED material impurity difference identification system is also used for: constructing a sliding analysis window for the aligned chromatographic sequence along the retention time direction; extracting a set of local minimum points of characteristic signal intensity within each sliding analysis window; generating an initial baseline control node sequence based on the set of local minimum points; constructing an initial baseline estimation curve by piecewise interpolation fitting of the initial baseline control node sequence; performing baseline stripping processing on the aligned chromatographic sequence with the initial baseline estimation curve as a constrained envelope to obtain residual signals; and calculating the signal based on the continuity of the residual signals between adjacent sliding analysis windows. Gradient change rate and curvature change rate are used to construct a continuity discrimination index characterizing the stability of the peak structure. Based on the continuity discrimination index, structural components are screened in the residual signal. Signal segments that simultaneously satisfy the amplitude continuity constraint, gradient monotonic change constraint, and curvature consistency constraint are identified as structural response components, and the remaining discrete high-frequency disturbance signals are eliminated as random noise components. Cross-batch scale uniform processing is performed on the structural response components. The amplitude of each batch of structural response components is normalized by the peak intensity ratio mapping relationship based on the cross-batch reference center spectrum, and signal contrast enhancement is performed in the local time neighborhood corresponding to each structural response component to form an impurity feature sequence.

[0083] Furthermore, the decomposition processing module 14 in the multi-batch OLED material impurity difference identification system is also used to: calculate the first-order gradient sequence and the second-order curvature sequence based on the residual signal within each sliding analysis window, and construct a continuity discrimination index in combination with the signal amplitude change. The continuity discrimination index is used to characterize the amplitude continuity, gradient change direction consistency and curvature change stability of the signal in the local time neighborhood.

[0084] Furthermore, the decomposition processing module 14 in the multi-batch OLED material impurity difference identification system is also used to: perform candidate structure segment division on the residual signal with the continuity discrimination index as a constraint, and identify the signal segment whose continuity discrimination index remains stable within a preset interval as a candidate structure segment; perform amplitude continuity constraint verification on the candidate structure segment to establish a first verification result; perform monotonicity interval division on the gradient sequence of the candidate structure segment to establish a second verification result; verify the curvature consistency of the peak region of the candidate structure segment based on the curvature sequence change trend to generate a third verification result; and establish a structural response component when the first verification result, the second verification result, and the third verification result are all passing results.

[0085] Furthermore, the decomposition processing module 14 in the multi-batch OLED material impurity difference identification system is also used to: perform cross-window consistency verification on the structural response components within adjacent sliding analysis windows; identify stable and drifting segments of the structural response components by comparing peak position offset, peak width change rate, and gradient direction consistency in adjacent sliding analysis windows; perform local time axis correction processing using the stability decay degree of the drifting segments; and perform cross-batch scale unification processing based on the correction processing results and the stable segments.

[0086] Furthermore, the anomaly determination module 15 in the multi-batch OLED material impurity difference identification system is also used to: read the multi-scale difference characterization quantity in the statistical characteristics of historical batches, construct a time series distribution model using the multi-scale difference characterization quantity; evaluate the degree of statistical distribution drift between different time batches using the time series distribution model, perform dynamic adaptive correction of the control threshold according to the evaluation result, and establish a dynamic control threshold.

[0087] Furthermore, the anomaly determination module 15 in the multi-batch OLED material impurity difference identification system is also used to: configure a batch anomaly identifier based on the batch impurity difference identification result, and use the batch anomaly identifier to perform anomaly diversion management of multiple batches of OLED materials.

[0088] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The method and specific examples for identifying impurity differences in multiple batches of OLED materials in the aforementioned embodiment 1 are also applicable to the system for identifying impurity differences in multiple batches of OLED materials in this embodiment. Through the foregoing detailed description of the method for identifying impurity differences in multiple batches of OLED materials, those skilled in the art can clearly understand the system for identifying impurity differences in multiple batches of OLED materials in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0089] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0090] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for identifying impurity differences in multiple batches of OLED materials, characterized in that, include: Acquire chromatographic detection data of multiple batches of OLED materials, perform unified sampling processing on the chromatographic detection data, and construct an original chromatographic sequence with retention time as the sequence index and response intensity as the numerical characterization; Based on the original chromatographic sequence, peak shape structural feature parameters are extracted from each batch of chromatographic data. The peak shape structural feature parameters include peak position, peak width position, peak shape asymmetry parameter and curvature change trend parameter. The corresponding morphology enhancement sequence is constructed using the peak shape structural feature parameters. After fusing the morphological enhancement sequence and the original chromatographic sequence, a cross-batch reference center spectrum is constructed, and nonlinear alignment processing of the original chromatographic sequence is performed under the constraint of the cross-batch reference center spectrum to generate an aligned chromatographic sequence. The process of fusing the morphological enhancement sequence and the original chromatographic sequence to construct a cross-batch reference center spectrum includes: The enhanced morphology sequences of each batch are weighted and fused with the corresponding original chromatographic sequences to form a fused sequence that highlights the peak shape and structural features. Based on the fusion sequence, the peak positions and peak structure intervals of each batch are extracted, and a set of peak structure correspondences across batches is constructed by matching and aligning the same peak structures in different batches. Under the constraint of the peak structure correspondence set, weighted superposition and position correction processing are performed on each batch of fused sequences within the corresponding peak structure interval, and smooth fusion is performed on non-peak structure intervals to generate a cross-batch reference center spectrum with unified peak structure distribution characteristics. Extract the retention times of all peaks in the original sequences of each batch, as well as the retention times of the corresponding peaks in the reference central spectrum, i.e., the fixed reference peaks. Use these two sets of time points as control point pairs, and construct a smooth time transformation function using piecewise polynomial interpolation. Map each retention time of the original sequence to a new retention time, so that the original peaks are mapped to the reference peaks. For regions without peak control points, the transformation function remains smooth and is as close to a linear transformation as possible to obtain the nonlinear aligned chromatographic sequence of that batch. Repeat the above steps for all batches to obtain the aligned chromatographic sequences of all batches. The aligned chromatographic sequence is input into the decomposition channel, and decomposition processing with adaptive baseline fitting and residual signal separation is performed to construct a pure impurity response signal after removing baseline interference. Normalization and local enhancement processing are then performed to form a cross-batch impurity characteristic sequence. A multi-scale difference characterization quantity is constructed based on the impurity feature sequence, and a dynamic control threshold is established by combining historical batch statistical features. The abnormal signal judgment of the multi-scale difference characterization quantity is performed using the dynamic control threshold, and the batch impurity difference identification result is output. The establishment of dynamic control thresholds based on historical batch statistical characteristics includes: Read the multi-scale difference characteristics from the historical batch statistical features, and use the multi-scale difference characteristics to construct a time series distribution model; The time series distribution model is used to evaluate the degree of statistical distribution drift between different time batches, and dynamic adaptive correction of the control threshold is performed based on the evaluation results to establish a dynamic control threshold.

2. The method for identifying impurity differences in multiple batches of OLED materials as described in claim 1, characterized in that, Input the aligned chromatographic sequence into the decomposition channel, including: A sliding analysis window is constructed along the retention time direction for the aligned chromatographic sequence. Within each sliding analysis window, a set of local minimum points of characteristic signal intensity is extracted, and an initial baseline control node sequence is generated based on the set of local minimum points. An initial baseline estimation curve is constructed by piecewise interpolation fitting of the initial baseline control node sequence. Using the initial baseline estimation curve as a constrained envelope, baseline stripping is performed on the aligned chromatographic sequence to obtain residual signals. Based on the continuity of the residual signals between adjacent sliding analysis windows, the rate of change of signal gradient and the rate of change of curvature are calculated to construct a continuity discrimination index characterizing the stability of peak structure. Based on the continuity discrimination index, the residual signal is screened for structural components. Signal segments that simultaneously satisfy the amplitude continuity constraint, gradient monotonic change constraint, and curvature consistency constraint are identified as structural response components, and the remaining discrete high-frequency disturbance signals are eliminated as random noise components. The structural response components are subjected to cross-batch scale uniform processing. The amplitude of each batch of structural response components is normalized by the peak intensity ratio mapping relationship based on the cross-batch reference center spectrum. Signal contrast enhancement is performed in the local time neighborhood corresponding to each structural response component to form an impurity feature sequence.

3. The method for identifying impurity differences in multiple batches of OLED materials as described in claim 2, characterized in that, Construct continuity criteria to characterize the stability of peak-shaped structures, including: Within each sliding analysis window, a first-order gradient sequence and a second-order curvature sequence are calculated based on the residual signal, and a continuity discrimination index is constructed by combining the signal amplitude change. The continuity discrimination index is used to characterize the amplitude continuity, gradient change direction consistency and curvature change stability of the signal in the local time neighborhood.

4. The method for identifying impurity differences in multiple batches of OLED materials as described in claim 3, characterized in that, Based on the aforementioned continuity discrimination index, structural component screening is performed on the residual signal, including: Using the continuity discrimination index as a constraint, the residual signal is divided into candidate structure segments, and signal segments whose continuity discrimination index remains stable within a preset interval are identified as candidate structure segments. Amplitude continuity constraint verification is performed on the candidate structural segments to establish a first verification result; The gradient sequence of the candidate structural fragments is divided into monotonic intervals to establish a second verification result; The curvature consistency of the peak region is verified based on the curvature sequence change trend of the candidate structural segments, and a third verification result is generated. If the first, second, and third verification results are all passed, then the structural response component is established.

5. The method for identifying impurity differences in multiple batches of OLED materials as described in claim 2, characterized in that, Before performing cross-batch scale uniform processing on the structural response components, the following is included: Cross-window consistency checks are performed on the structural response components within adjacent sliding analysis windows. By comparing the peak position offset, peak width change rate, and gradient direction consistency between adjacent sliding analysis windows, stable and drifting segments of the structural response components are identified. Local time axis correction is performed based on the stability decay of the drift segment, and cross-batch scale uniform processing is performed based on the correction results and the stable segment.

6. The method for identifying impurity differences in multiple batches of OLED materials as described in claim 1, characterized in that, Based on the original chromatographic sequence, peak shape and structural characteristic parameters were extracted from each batch of chromatographic data, including: The first-order gradient sequence and the second-order curvature sequence are calculated along the retention time direction of the original chromatographic sequence, and the candidate intervals of the chromatographic signal are divided according to the position of gradient sign change to determine the time boundary of each peak structure. Within each candidate peak region, the peak position is determined based on the maximum response intensity of the peak region, and the peak width position is determined based on the time span corresponding to the signal attenuation to a preset ratio threshold on both sides of the peak position. Peak shape asymmetry parameters are constructed based on the gradient integral difference on both sides of the peak position. The peak position, peak width, peak shape asymmetry parameter, and curvature change trend parameter are coupled and fused. The original chromatographic sequence is then locally weighted and reconstructed using the coupling and fusion results to obtain a morphologically enhanced sequence.

7. The method for identifying impurity differences in multiple batches of OLED materials as described in claim 1, characterized in that, Configure batch anomaly identifiers based on batch impurity difference identification results, and use the batch anomaly identifiers to perform anomaly diversion management of multiple batches of OLED materials.

8. A system for identifying impurity differences in multiple batches of OLED materials, characterized in that, The step of implementing the multi-batch OLED material impurity difference identification method according to any one of claims 1 to 7, wherein the multi-batch OLED material impurity difference identification system comprises: A unified sampling and processing module is used to acquire chromatographic detection data of multiple batches of OLED materials, perform unified sampling and processing of the chromatographic detection data, and construct an original chromatographic sequence with retention time as the sequence index and response intensity as the numerical characterization. The feature parameter extraction module is used to extract peak shape structure feature parameters from each batch of chromatographic data based on the original chromatographic sequence. The peak shape structure feature parameters include peak position, peak width position, peak shape asymmetry parameter and curvature change trend parameter, and construct the corresponding morphology enhancement sequence using the peak shape structure feature parameters. The reference center spectrum construction module is used to construct a cross-batch reference center spectrum after fusing the morphological enhancement sequence and the original chromatographic sequence, and to perform nonlinear alignment processing of the original chromatographic sequence under the constraint of the cross-batch reference center spectrum to generate an aligned chromatographic sequence. The reference center spectrum construction module is further used to: weightedly fuse the morphological enhancement sequences of each batch with the corresponding original chromatographic sequences to form a fusion sequence that highlights the peak shape and structural features; extract the corresponding peak positions and peak structure intervals of each batch based on the fusion sequence, and construct a set of peak structure correspondences across batches by matching and aligning the same peak structures in different batches; under the constraints of the set of peak structure correspondences, perform weighted superposition and position correction processing on the fusion sequences of each batch within the corresponding peak structure intervals, and perform smooth fusion on non-peak structure intervals to generate a cross-batch reference center spectrum with uniform peak structure distribution features; The system is also used to extract the retention times of all peaks in the original sequences of each batch, as well as the retention times of the corresponding peaks in the reference center spectrum, i.e., the fixed reference peaks. Using these two sets of time points as control point pairs, a smooth time transformation function is constructed using piecewise polynomial interpolation to map each retention time of the original sequence to a new retention time, so that the original peaks are mapped to the reference peaks. For regions without peak control points, the transformation function remains smooth and is as close to a linear transformation as possible to obtain the nonlinear aligned chromatographic sequence of that batch. The above steps are repeated for all batches to obtain the aligned chromatographic sequences of all batches. The decomposition processing module is used to input the aligned chromatographic sequence into the decomposition channel, perform decomposition processing of adaptive baseline fitting and residual signal separation, construct a pure impurity response signal after removing baseline interference, and perform normalization and local enhancement processing to form impurity characteristic sequences across batches. The anomaly detection module is used to construct a multi-scale difference characterization quantity based on the impurity feature sequence, and to establish a dynamic control threshold by combining historical batch statistical features. The dynamic control threshold is used to perform anomaly signal detection of the multi-scale difference characterization quantity and output batch impurity difference identification results. The anomaly detection module is also used to read the multi-scale difference representation in the statistical features of historical batches, construct a time series distribution model using the multi-scale difference representation, evaluate the degree of statistical distribution drift between different time batches using the time series distribution model, perform dynamic adaptive correction of the control threshold based on the evaluation results, and establish a dynamic control threshold.

Images