Centrifugal detection system

Abstract

The invention relates to the technical field of centrifugal detection, and provides a centrifugal detection system which is characterized in that spectral data, image data and pressure data are acquired through a detection module, spectral features, layered edge sharpness and pressure change features are extracted, the features are fused through an attention mechanism to obtain centrifugal fusion features, and then an accurate centrifugal state score is analyzed; the control module combines the centrifugal state score and simulates the vibration amplitude and torque change in the centrifugal detection process to judge whether the centrifugal state is abnormal or not, the centrifugal state can be comprehensively and accurately evaluated in a multi-source data detection and multi-factor comprehensive judgment mode, the limitation of single data or single factor judgment is avoided, and the detection accuracy is improved. Potential problems of physical and biological levels in the centrifugal process can be timely and accurately found, and a reliable basis is provided for subsequent adjustment and control.

Description

Technical Field

[0001] This application relates to the field of centrifugal detection technology, and particularly to a centrifugal detection system. Background Art

[0002] In the fields of medical treatment and scientific research, the analysis and detection of samples are crucial. The quality of samples directly affects the effects of subsequent research and treatment. Existing centrifugal detection systems often rely on only a single type of data for analysis and cannot comprehensively reflect the state of samples during the simulated centrifugation process. Moreover, when judging whether the centrifugation state is abnormal, usually only a single factor is considered, ignoring the influence of other important factors on the centrifugation state, resulting in inaccurate abnormal judgment, inability to detect potential problems in a timely manner, thus affecting the normal operation of the simulated centrifugation detection process and the quality of samples, and lacking targeted adjustment, leading to further decline in sample quality, and not solving the problem of how to achieve improved centrifugal detection of the target sample quality. Summary of the Invention

[0003] In view of the deficiencies of the prior art, this application provides a centrifugal detection system, including: a detection module and a control module;

[0004] The detection module is used to obtain spectral data, image data, and pressure data, extract spectral features, stratification edge sharpness, and pressure change features, perform feature fusion on the spectral features, stratification edge sharpness, and pressure change features through an attention mechanism to obtain a centrifugal fusion feature, and analyze the centrifugal fusion feature to obtain a centrifugation state score;

[0005] The control module is used to monitor the vibration amplitude and torque change during the simulated centrifugation detection process, and combine the centrifugation state score to judge whether the centrifugation state is abnormal. If the centrifugation state is abnormal, the centrifugation parameters are adjusted, and it is judged whether to trigger sample distribution adjustment and inertia deviation compensation.

[0006] As an optional implementation manner, the acquisition logic of the spectral data includes:

[0007] Integrate a dual-wavelength laser and trigger the dual-wavelength laser alternately in time division;

[0008] Start the acquisition of spectral data when the target sample is in the detection window;

[0009] Analyze the spectral features of the dual-wavelength laser channels in the spectral data respectively.

[0010] As an optional implementation manner, the image data is obtained through an image sensor, and the pressure data is obtained through a pressure sensor. The extraction logic of the stratification edge sharpness includes:

[0011] Identify the edge pixels of the image data through an edge algorithm, and segment the image data into multiple stratification regions according to the edge pixels of the image data;

[0012] Configure a local window and calculate the gradient magnitude and gradient direction of the edge pixels in each hierarchical region within the local window;

[0013] Comprehensively analyze the gradient magnitude, gradient direction, and local contrast of the edge pixels to obtain the hierarchical edge sharpness.

[0014] As an optional implementation manner, the segmentation sub-logic of the hierarchical region includes:

[0015] Generate an adjacency graph based on the edge pixels of the image data, where each node represents an edge pixel and the edge represents the connection between adjacent edge pixels;

[0016] Calculate the similarity between nodes and configure a similarity threshold;

[0017] Cluster the nodes with similarity greater than the similarity threshold into the same hierarchical region through a clustering algorithm.

[0018] As an optional implementation manner, the logic of feature fusion includes:

[0019] Construct a spatio-temporal matrix of spectral features, hierarchical edge sharpness, and pressure features, and calculate the spatial attention weight according to the spatio-temporal matrix;

[0020] Calculate the temporal attention weight according to the spatio-temporal matrix;

[0021] Perform a Hadamard product on the spatial attention weight and the temporal attention weight to obtain the centrifugal fusion feature.

[0022] As an optional implementation manner, the judgment logic for whether the centrifugal state is abnormal includes:

[0023] Determine the physical abnormal state according to the vibration amplitude and torque change during the simulated centrifugal detection process;

[0024] Configure a scoring threshold and compare the centrifugal state score with the scoring threshold to obtain the biological abnormal state;

[0025] Comprehensively judge whether the centrifugal state is abnormal according to the physical abnormal state and the biological abnormal state.

[0026] As an optional implementation manner, the judgment logic for triggering the adjustment of the sample distribution includes:

[0027] Monitor the pressure data in real time to determine the adjustment trigger condition for the sample distribution;

[0028] Generate an annular pressure gradient map based on the pressure data, mark the pressure distribution regions, where the pressure distribution regions include regions of sudden pressure increase and regions of sudden pressure drop, and judge the imbalance direction;

[0029] Determine the air flow type according to the pressure distribution area, and determine the air flow intensity according to the imbalance direction;

[0030] Update the annular pressure gradient map in real time to verify the adjustment effect.

[0031] As an optional implementation manner, the imbalance direction is judged according to the distribution mode of the pressure sudden increase area and the pressure sudden drop area. If the pressure sudden increase area and the pressure sudden drop area are symmetrically distributed, it is judged as static imbalance;

[0032] If the pressure sudden increase area and the pressure sudden drop area are asymmetrically distributed, it is judged as dynamic imbalance.

[0033] As an optional implementation manner, the judgment logic for triggering inertia deviation compensation includes:

[0034] Monitor the sharpness of the stratified edge in real time to determine the compensation trigger condition for inertia deviation compensation;

[0035] Monitor the torque change to judge the authenticity of the compensation trigger condition;

[0036] If the compensation trigger condition is true, adjust the rotational speed in segments to compensate for the inertia deviation;

[0037] Identify the resonance state during the process of adjusting the rotational speed in segments to execute the avoidance mechanism;

[0038] Continuously monitor the change of the sharpness of the stratified edge to verify the compensation effect.

[0039] As an optional implementation manner, the sub-logic of adjusting the rotational speed in segments includes:

[0040] When the compensation trigger condition is true, reduce the rotational speed;

[0041] During the process of reducing the rotational speed, monitor the decrease amplitude of the torque change. If the decrease amplitude of the torque change is greater than the amplitude threshold, restore the rotational speed to the target rotational speed;

[0042] If the decrease amplitude of the torque change is less than or equal to the amplitude threshold, identify the resonance state to execute the avoidance mechanism.

[0043] Compared with the prior art, the beneficial effects of the present application are as follows: The detection module acquires spectral data, image data, and pressure data, extracts spectral features, layered edge sharpness, and pressure change features, and fuses these features through an attention mechanism to obtain a centrifugal fusion feature, and then analyzes an accurate centrifugal state score; the control module combines this centrifugal state score and the vibration amplitude and torque changes during the simulated centrifugal detection process to determine whether the centrifugal state is abnormal. This method of multi-source data detection and multi-factor comprehensive judgment can comprehensively and accurately evaluate the centrifugal state, avoid the limitations of single-data or single-factor judgment, and can timely and accurately discover potential problems at the physical and biological levels during the centrifugation process, providing a reliable basis for subsequent adjustment and control.

[0044] The detection module can acquire spectral data, image data, and pressure data, and extract spectral features, layered edge sharpness, and pressure change features from them, enabling a comprehensive and detailed understanding of the state changes of the sample during centrifugation, providing rich information for accurately evaluating the centrifugal state; at the same time, through the attention mechanism, the spectral features, layered edge sharpness, and pressure change features are feature-fused to obtain a centrifugal fusion feature, which can give full play to the advantages of multi-source features, highlight key information, and analyze and obtain a centrifugal state score, which can comprehensively consider the influence of various factors on the centrifugal state, helping the operator to timely understand the operating conditions of the centrifugal detection system and the centrifugation effect of the sample, and providing a strong basis for subsequent operations and decisions.

[0045] The control module monitors the vibration amplitude and torque changes during the simulated centrifugal detection process, and combines the centrifugal state score to determine whether the centrifugal state is abnormal. Considering factors at both the physical and biological levels, it can more comprehensively and accurately discover abnormal situations during centrifugation, avoiding centrifuge failures and sample quality degradation; when it is determined that the centrifugal state is abnormal, the control module can determine whether to trigger sample distribution adjustment and inertia deviation compensation. By real-time monitoring of pressure data and layered edge sharpness, it can timely discover uneven sample distribution and inertia deviation, and take corresponding adjustment and compensation measures; among them, the sample distribution adjustment mechanism can use the air pressure gradient to adjust the sample distribution to make the sample distribution more uniform, and the inertia deviation compensation mechanism can compensate for the inertia deviation by means of segmented speed adjustment, etc., to avoid the occurrence of resonance phenomena, ensuring the stable operation of the centrifugal detection system and the high-quality preparation of samples. Description of the Drawings

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. Among them:

[0047] Figure 1 It is a system flowchart of a centrifugal detection system provided by an embodiment of the present application;

[0048] Figure 2 It is a sub-logic diagram for dividing the layered area of a centrifugal detection system provided by an embodiment of the present application;

[0049] Figure 3 It is a feature fusion logic diagram of a centrifugal detection system provided by an embodiment of the present application;

[0050] Figure 4 It is a judgment logic diagram for segmentally adjusting the rotation speed of a centrifugal detection system provided by an embodiment of the present application. Specific embodiments

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present application more obvious and understandable, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings of the specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.

[0052] As Figure 1 shown, a system flowchart of a centrifugal detection system is provided by an embodiment of the present application. A centrifugal detection system includes a detection module and a control module.

[0053] The target sample in this embodiment is exemplified as a PRP sample.

[0054] The detection module is used to obtain spectral data, image data, and pressure data, extract spectral features, stratification edge sharpness, and pressure change features, perform feature fusion on the spectral features, stratification edge sharpness, and pressure change features through an attention mechanism to obtain centrifugal fusion features, and analyze the centrifugal fusion features to obtain a centrifugal state score.

[0055] Specifically, the acquisition logic of spectral data includes:

[0056] Integrate a dual-wavelength laser and alternately trigger the dual-wavelength laser at different times;

[0057] When the PRP sample is in the detection window, start the acquisition of spectral data;

[0058] Parse the spectral features of the dual-wavelength laser channels in the spectral data respectively.

[0059] To better detect the PRP sample, it is necessary to obtain spectral data, image data, and pressure data here. The image data is obtained by photographing the stratification interface in the PRP sample container through an image sensor (such as a camera), and the pressure data is obtained by monitoring the pressure change in the PRP sample container through a pressure sensor. The acquisition logic of spectral data is described in detail here, and the acquisition logics of other data will not be elaborated one by one.

[0060] The composition of the PRP sample is complex, and different components have different absorption characteristics for light of different wavelengths. Dual-wavelength lasers can cover more component information, and time-division alternating triggering can avoid interference and improve data accuracy. Two specific wavelength laser generators that are sensitive to the key components of the PRP sample are selected. For example, the wavelengths are selected according to the absorption peaks of the characteristic substances in platelets and plasma. Suppose the wavelengths are λ1 and λ2 respectively. In the spectral data acquisition stage, the dual-wavelength lasers are triggered alternately in time division according to the preset trigger timing logic. For example, in the first half of a preset rotation period, the laser generator with wavelength λ1 is triggered; in the second half, the laser generator with wavelength λ2 is triggered to avoid interference between lasers of different wavelengths and achieve time-division alternation, so as to obtain spectral data that more purely and accurately reflects the composition of the PRP sample, providing a high-quality data basis for subsequent spectral feature analysis. Accurate spectral data can make the extracted spectral features more representative.

[0061] To ensure that only the spectral data of the PRP sample in the PRP sample container is obtained, avoid obtaining invalid background information, and improve the effectiveness of the spectral data, the detection window can be set as a 30° oblique angle area of the axis of the PRP sample container. At the same time, a photoelectric sensor is set at the position of the detection window. Through the collaborative work of the position sensor and the photoelectric sensor, when the PRP sample enters the detection window, it will block the light and cause a level change signal in the photoelectric sensor. Analyze the signals of the position sensor and the photoelectric sensor to determine the moment when the PRP sample is in the detection window, and then start the acquisition of spectral data, so as to reduce the acquisition of invalid data and improve the acquisition efficiency and data pertinence.

[0062] The spectral data of different wavelength laser channels reflects different aspects of information of the PRP sample. Separately analyzing them helps to comprehensively understand the composition and state of the PRP sample. The spectral data of each wavelength channel is processed through Fourier transform to convert the spectral data in the time domain into frequency domain data, so as to extract the parameters of the spectral features, such as the position, intensity, and full width at half maximum of the characteristic peaks, etc. For complex spectral data, combined with wavelet analysis, the spectral data is decomposed into different scale spaces to further explore the detailed information of the spectral data and enhance the recognition ability of weak characteristic peaks, so as to obtain detailed sample composition feature information and provide rich spectral feature dimensions for subsequent feature fusion.

[0063] After obtaining spectral data, image data, and pressure data, it is necessary to extract the features of the data, including spectral features, stratification edge sharpness, and pressure change features. The extraction logic of the stratification edge sharpness is described in detail here, and the extraction logics of other features will not be elaborated one by one.

[0064] Specifically, the extraction logic of the stratification edge sharpness includes:

[0065] Identify the edge pixels of the image data through an edge algorithm, and segment the image data into several hierarchical regions according to the edge pixels of the image data;

[0066] Configure a local window, and calculate the gradient magnitude and gradient direction of the edge pixels of each hierarchical region within the local window;

[0067] Comprehensively analyze the gradient magnitude, gradient direction, and local contrast of the edge pixels to obtain the hierarchical edge sharpness.

[0068] After centrifugation, the PRP sample is stratified, and the edge features can reflect the stratification situation. Segmenting the stratified regions facilitates targeted analysis of the characteristics of each layer. Identify the edge pixels of the image data through the Canny edge detection algorithm. First, perform Gaussian filtering on the acquired image data to remove noise interference; then calculate the gradient magnitude and direction of each pixel in the image data; next, refine the edges through non-maximum suppression; finally, use double-threshold detection and edge tracking to determine the final edge pixels; based on the spatial distribution and connection relationship of the edge pixels, segment the image data into different hierarchical regions in the form of an adjacency graph, so as to clearly divide the stratification structure of the sample and provide an accurate region division for subsequent calculation of the hierarchical edge sharpness.

[0069] Analyzing the characteristics of the edge pixels locally can more meticulously reflect the detailed information of the hierarchical edges and provide basic data for comprehensive analysis. Set a square or rectangular local window of an appropriate size according to the resolution of the image data and the size of the hierarchical region. For example, for an image with a resolution of 1024×768, if the size of the hierarchical region is between dozens and hundreds of pixels, set the window size to 3×3, 5×5, or 7×7, etc.; within each local window, perform a convolution operation on the edge pixels through the Sobel operator to obtain the gradient magnitude and gradient direction of the edge pixels. The Sobel operator convolves the image data horizontally and vertically through two 3×3 convolution kernels respectively, and then calculates the gradient magnitude and gradient direction according to relevant formulas, so as to obtain the detailed gradient information of the edge pixels of each hierarchical region and reflect the change trend and intensity of the edge pixels.

[0070] A single feature cannot comprehensively reflect the clarity and stability of the hierarchical edges. Combining multiple features can more accurately quantify the hierarchical edge sharpness. Therefore, different weights are assigned to the gradient magnitude, gradient direction, and local contrast, and the hierarchical edge sharpness is calculated through a weighted summation formula. The weights are optimized and determined through a machine learning algorithm. For example, through the support vector regression algorithm, the gradient magnitude, gradient direction, and local contrast of a large number of samples with known hierarchical quality are used as input features, and the manually labeled hierarchical edge sharpness is used as the output label to train the support vector regression model, so as to obtain the optimal weight assignment, and thus obtain a quantitative index that can accurately reflect the stratification quality of the PRP sample, providing an important basis for judging the centrifugation state.

[0071] Furthermore, as Figure 2 shown, the segmentation sub-logic of the hierarchical region includes:

[0072] Generate an adjacency graph based on the edge pixels of the image data, where each node represents an edge pixel and the edges represent the connections between adjacent edge pixels;

[0073] Calculate the similarity between nodes and configure a similarity threshold;

[0074] Group the nodes with similarity greater than the similarity threshold into the same hierarchical region through a clustering algorithm.

[0075] Convert the spatial relationship of the edge pixels of the image data into a graph structure, which is convenient for using graph theory algorithms for region segmentation. Traverse the edge pixels of the image data, create nodes for each edge pixel, determine adjacent edge pixels according to the spatial position relationship of the edge pixels, and establish edges between the nodes to construct an adjacency graph. For example, for the edge pixel (i, j) in a two-dimensional image, its adjacent pixels include (i - 1, j), (i + 1, j), (i, j - 1), and (i, j + 1). The number of adjacent pixels of the boundary pixels will be reduced. Establish edges between the corresponding nodes and use an adjacency matrix or adjacency list to store the adjacency graph, which is convenient for subsequent graph operations, so as to represent the edge pixel relationship with an intuitive graph structure and provide a data structure basis for subsequent clustering analysis.

[0076] Measure the similarity between nodes through similarity, and determine which nodes should belong to the same region. The similarity threshold is used to control the tightness of clustering. Calculate the similarity between nodes through features such as the gray value, gradient magnitude, and gradient direction of each node. For example, for two nodes A and B, their feature vectors are [a1, a2, a3] and [b1, b2, b3] respectively. a1 represents the gray value of node A, a2 represents the gradient magnitude of node A, a3 represents the gradient direction of node A, b1 represents the gray value of node B, b2 represents the gradient magnitude of node B, and b3 represents the gradient direction of node B. Then the calculation formula for the similarity d between node A and node B is: Determine a suitable similarity threshold through cross-validation, so as to provide a quantitative basis for clustering and decide which nodes will be merged into the same hierarchical region.

[0077] To achieve automatic grouping of edge pixels and form different hierarchical regions, the K-Means clustering algorithm is selected. According to the calculated similarity between nodes and the set similarity threshold, the nodes in the adjacency graph are clustered into different hierarchical regions. For example, first randomly select K initial clustering centers, and then calculate the similarity between each node and each clustering center. Here, the similarity can be calculated by the Euclidean distance, and the node is assigned to the cluster where the clustering center with the highest similarity is located; then recalculate the clustering center of each cluster, and repeat the above process until the clustering center no longer changes or reaches the preset number of iterations, so as to accurately divide the hierarchical region of the PRP sample and improve the accuracy of calculating the hierarchical edge sharpness.

[0078] Specifically, as Figure 3 shown, the logic of feature fusion includes:

[0079] Construct the spatio-temporal matrices of spectral features, hierarchical edge sharpness, and pressure features, and calculate the spatial attention weights according to the spatio-temporal matrices;

[0080] Calculate the temporal attention weights according to the spatio-temporal matrices;

[0081] Perform the Hadamard product on the spatial attention weights and the temporal attention weights to obtain the centrifugal fusion features.

[0082] Different features have different effects on the centrifugal state in the spatial and temporal dimensions. Constructing the spatio-temporal matrix can uniformly represent multi-source features. Calculating the spatial attention weights can highlight the important features in space. Arrange the spectral features, hierarchical edge sharpness, and pressure features in chronological order to construct a three-dimensional spatio-temporal matrix. For example, assume that the dimension of the spectral feature is m, the dimension of the hierarchical edge sharpness is n, the dimension of the pressure feature is p, and the time step is t. Then the dimension of the spatio-temporal matrix is m×n×p×t. Perform convolution operations on the spatio-temporal matrix through a convolutional neural network, and calculate the spatial attention weights through the Softmax function.

[0083] Specifically, design multiple convolutional layers. Each convolutional layer contains multiple convolutional kernels. By sliding the convolutional kernels on the spatio-temporal matrix, extract spatial features of different scales. Then pass the output of the convolutional layer through a fully connected layer, and then normalize it through the Softmax function to obtain the matrix of spatial attention weights, so as to highlight the importance of different features in the spatial dimension and make the fusion features pay more attention to the key spatial information.

[0084] Considering the changing trend of features in the time dimension, calculating the time attention weights can emphasize the feature information at important time points. The spatio-temporal matrix is processed through a long short-term memory network. Through the gating mechanism and weight calculation, the time attention weights are obtained. The LSTM unit includes an input gate, a forget gate, and an output gate. Through the control of these gates, the information in the time series is selectively memorized and updated. The spatio-temporal matrix is unfolded along the time dimension and sequentially input into the long short-term memory network. The long short-term memory network calculates the weights at each time step according to the input information and internal state, and finally normalizes through the Softmax function to obtain the matrix of time attention weights, thereby capturing the changing trend of features in the time dimension and highlighting the features at key time points.

[0085] Combining the attention weights in the spatial and time dimensions to achieve the effective fusion of multi-source features and obtain a comprehensive feature that can better reflect the centrifugal state. Multiply the corresponding elements of the matrix of spatial attention weights and the matrix of time attention weights to obtain the matrix of centrifugal fusion features. For example, if the matrix of spatial attention weights is S, the matrix of time attention weights is T, and the matrix of centrifugal fusion features is F, then F pq = S pq × T pq , where p and q respectively represent the row and column indices of the matrix, thereby effectively fusing multi-source features and generating a comprehensive feature that comprehensively reflects the centrifugal state.

[0086] After normalizing the centrifugal fusion features, input them into a multi-layer perceptron for classification and regression analysis. The multi-layer perceptron consists of an input layer, multiple hidden layers, and an output layer. The number of nodes in the input layer is the same as the dimension of the matrix of centrifugal fusion features. The hidden layer can be set to 2 - 3 layers, and the number of neurons in each layer is determined according to experiments, such as selecting from values such as 64, 128, and 256. The number of nodes in the output layer is the same as the number of centrifugal state indicators to be quantified. For example, if the quantified indicators of the centrifugal state are stratification uniformity, sample separation efficiency, and centrifugal stability, then the number of node data in the output layer is 3. Then, configure a weight for stratification uniformity, sample separation efficiency, and centrifugal stability respectively, and calculate the centrifugal state score through weighted summation.

[0087] The control module is used to monitor the vibration amplitude and torque change during the simulated centrifugal detection process, and combine the centrifugal state score to judge whether the centrifugal state is abnormal. If the centrifugal state is abnormal, adjust the centrifugal parameters and judge whether to trigger sample distribution adjustment and inertia deviation compensation.

[0088] The centrifugal parameters include rotational speed, centrifugal time, and temperature;

[0089] Specifically, the judgment logic for whether the centrifugal state is abnormal includes:

[0090] Determine the physical abnormal state based on the vibration amplitude and torque change during the simulated centrifugation detection process;

[0091] Configure a scoring threshold, and compare the centrifugation state score with the scoring threshold to obtain the biological abnormal state;

[0092] Comprehensively judge whether the centrifugation state is abnormal according to the physical abnormal state and the biological abnormal state.

[0093] During the process of simulated centrifugation, its vibration amplitude and torque will be maintained within a relatively stable range. When physical parameter abnormalities such as uneven sample distribution occur, it will directly cause abnormal fluctuations in the vibration amplitude and torque. By monitoring these two key parameters, the abnormal operation at the physical level during the simulated centrifugation process can be detected in a timely manner, providing a basis for subsequent corresponding measures; among them, the vibration amplitude is monitored through an acceleration sensor, and the torque change is monitored through a torque sensor to obtain the vibration amplitude and torque change in the simulation system respectively; the analog signals output by the acceleration sensor and the torque sensor are amplified and filtered, and the conditioned signals are converted into digital signals, including vibration signals and torque signals. The vibration signal is subjected to spectral analysis through the Fourier transform algorithm to extract the amplitude of the vibration as the vibration amplitude; at the same time, the torque signal is subjected to time-domain analysis, and the standard deviation of the torque is calculated as the torque change. The vibration amplitude and the torque change are respectively compared with the preset amplitude threshold and change threshold. If the vibration amplitude is greater than the preset amplitude threshold or the torque change is greater than the preset change threshold, it is determined as the physical abnormal state, so that the abnormal changes in the physical operation process during the simulated centrifugation process can be captured in real time and accurately, providing a reliable means for timely discovery of potential faults, and effectively ensuring the stability of the simulation environment and the reliability of the detection results.

[0094] The centrifugation state score is obtained by integrating multi-source information such as spectral data, image data, and pressure data. It can reflect the biological characteristics and state of the sample during centrifugation. By setting a reasonable scoring threshold and comparing it with the actual centrifugation state score, it can be judged whether the centrifugation effect of the sample is normal from the biological level, providing supplementary information for comprehensively evaluating the centrifugation state. By statistically analyzing the historical centrifugation state scores, a suitable scoring threshold is determined. If the centrifugation state score is less than the scoring threshold, it is determined as the biological abnormal state, quantitatively evaluating the centrifugation state from the perspective of biological characteristics, providing a scientific basis for judging the quality of the sample and the centrifugation effect, and improving the ability to identify abnormal situations during the centrifugation process.

[0095] Neither the physical abnormal state nor the biological abnormal state alone can comprehensively and accurately evaluate the centrifugation state. Physical abnormalities will affect the separation effect of the sample, while biological abnormalities directly reflect the state of the sample itself. Considering both physical and biological abnormalities comprehensively can better grasp the operating conditions during the simulated centrifugation process and the centrifugation effect of the sample, avoiding misjudgment caused by single-factor judgment. Only when both physical and biological abnormalities exist can it be determined that the centrifugation state is abnormal, thereby improving the accuracy and reliability of the abnormal centrifugation state judgment, being able to more comprehensively discover potential problems in the simulated centrifugation process, and providing strong support for taking timely measures.

[0096] Rotational speed is one of the key parameters affecting the centrifugation effect. When the centrifugation state is abnormal, appropriately adjusting the rotational speed can change the centrifugal force on the sample, thereby improving the separation effect of the sample and restoring the normal operation of the simulated centrifugation process. Among them, if the sample stratification is unclear due to too high a rotational speed, the rotational speed is reduced; if the separation time is too long due to too low a rotational speed, the rotational speed is increased, so that the centrifugal rotational speed can be flexibly and accurately adjusted according to the actual situation, effectively improving the centrifugation effect of the sample, solving the problem of abnormal centrifugation state caused by improper rotational speed, and improving the efficiency of the simulated detection and the quality of the sample simulation analysis. The simulated adjustment of the centrifugal rotational speed is achieved by changing the frequency of the input voltage of the simulated motor, thereby precisely adjusting the centrifugal rotational speed.

[0097] The length of the centrifugation time directly affects the separation degree of the sample. When the centrifugation state is abnormal, if the sample is not separated sufficiently or over-separated due to insufficient or too long centrifugation time, adjusting the centrifugation time can allow the sample to have enough time for separation, thereby improving the quality of the sample, ensuring that the sample completes the centrifugation process within an appropriate time, optimizing the separation effect of the sample, avoiding sample quality problems caused by improper centrifugation time, and improving the working efficiency and reliability of the simulated centrifugation detection.

[0098] Temperature has an important impact on the activity and physical and chemical properties of the sample. During centrifugation, too high or too low a temperature will cause the quality of the sample to decline. Adjusting the temperature can provide a suitable environment for the sample, ensuring the activity and stability of the sample, thereby improving the centrifugation effect, reducing the situation of poor centrifugation effect caused by temperature problems, and improving the reliability of the simulated centrifugation detection and the usability of the sample simulation analysis.

[0099] Specifically, the judgment logic for triggering sample distribution adjustment includes:

[0100] Real-time monitor the pressure data to determine the adjustment trigger condition for sample distribution adjustment;

[0101] Generate an annular pressure gradient map based on the pressure data, mark the pressure distribution area, the pressure distribution area includes the pressure sudden increase area and the pressure sudden drop area, and judge the imbalance direction;

[0102] Determine the air flow type based on the pressure distribution area, and determine the air flow intensity based on the imbalance direction;

[0103] Update the annular pressure gradient map in real time to verify the adjustment effect.

[0104] After adjusting the centrifugal parameters, monitor the pressure data according to the sampling period to timely detect problems with the sample distribution, provide an accurate basis for triggering sample distribution adjustment, ensure the stable operation of the simulated centrifugal detection and the quality of the samples, calculate the standard deviation of the pressure data, analyze the pressure change over time, and set a pressure change threshold. When the pressure change of the monitored pressure data is greater than the pressure change threshold, it is determined that the adjustment trigger condition is met, so as to be able to monitor the pressure change in the simulated centrifugal environment in real time and accurately, timely detect the problem of uneven sample distribution, provide a reliable trigger signal for subsequent sample distribution adjustment operations, and improve the operation stability of the simulated centrifugal detection and the sample quality.

[0105] The annular pressure gradient map can intuitively display the pressure distribution in the simulated centrifugal environment. By marking the pressure distribution area and judging the imbalance direction, the specific situation of uneven sample distribution can be understood more accurately, providing a visual basis for subsequent determination of sample distribution adjustment. Process the obtained discrete pressure data, convert the discrete data points into continuous pressure distribution data through bilinear interpolation, and then map these pressure distribution data onto an annular coordinate system to generate an annular pressure gradient map; divide the pressure distribution area into a pressure sudden increase area and a pressure sudden drop area. For example, the pressure sudden increase area refers to the area where the pressure change is greater than 1.5 times the pressure change threshold for two consecutive sampling periods, while the pressure sudden drop area refers to the area where there are more than two pressure sensors and the pressure change is less than 0.5 times the pressure change threshold. At the same time, in the annular pressure gradient map, use different colors or marks to distinguish these areas for intuitive observation of the pressure distribution.

[0106] Furthermore, the imbalance direction is judged according to the distribution pattern of the pressure sudden increase area and the pressure sudden drop area. If the pressure sudden increase area and the pressure sudden drop area are symmetrically distributed, it is judged as static imbalance;

[0107] If the pressure sudden increase area and the pressure sudden drop area are asymmetrically distributed, it is judged as dynamic imbalance.

[0108] Judge the imbalance direction according to the distribution mode of the pressure sudden increase area and the pressure sudden drop area. If the pressure sudden increase area and the pressure sudden drop area are symmetrically distributed (180°), it is judged as static imbalance; if the pressure sudden increase area and the pressure sudden drop area are asymmetrically distributed, such as 120°, it is judged as dynamic imbalance, so as to intuitively present the pressure distribution in the simulated centrifugal environment and the sample imbalance situation, providing a clear and visual basis for determining the air flow type and intensity, and helping to improve the accuracy and efficiency of sample distribution adjustment.

[0109] Different pressure distribution areas and imbalance directions require different air flow types and intensities to adjust the sample distribution. Determining the appropriate air flow type according to the actual situation can make more effective use of the air pressure gradient to adjust the sample distribution and improve the problem of uneven sample distribution; for the pressure sudden increase area, adsorb through negative pressure air flow to reduce the pressure in this area; for the pressure sudden drop area, fill it with positive pressure air flow to increase the pressure in this area, and form an appropriate air pressure gradient by controlling the direction and flow rate of the air flow to promote the redistribution of the sample.

[0110] Determine the air flow intensity according to the imbalance direction and the degree of pressure change. For example, for the case of dynamic imbalance and large pressure change, appropriately increase the air flow intensity; for the case of static imbalance and small pressure change, reduce the air flow intensity. Among them, the corresponding type and intensity of air flow can be generated by the simulated air flow device, and the flow rate of the air flow can be controlled by adjusting the opening of the simulated valve in the simulated air flow device to ensure that the air flow intensity meets the requirements. Therefore, adjusting the air flow type and intensity targeted can make more effective use of the simulated air pressure gradient to adjust the sample distribution, make the sample distribution more uniform, and improve the operation stability of the simulated centrifugal detection and the separation effect of the sample.

[0111] During the process of sample distribution adjustment, real-time updating of the annular pressure gradient map can intuitively observe the change of the pressure distribution, so as to verify the adjustment effect and judge whether it is necessary to further adjust the air flow type and air flow intensity to ensure that the sample distribution adjustment reaches the expected goal. Continuously obtain the data of the pressure sensor to update the annular pressure gradient map in real time, compare the annular pressure gradient maps before and after adjustment, observe the change of the pressure distribution area and the imbalance direction, and calculate the uniformity index of the pressure distribution, such as the variance of the pressure data, to evaluate the adjustment effect; if the pressure distribution gradually tends to be uniform and the imbalance situation is improved, it means that the adjustment effect is good; otherwise, it is necessary to further adjust the air flow type or intensity, so as to timely feedback the effect of the sample distribution adjustment, provide a basis for the optimization of the sample distribution adjustment, ensure that the sample distribution adjustment reaches the expected goal, and improve the operation efficiency of the simulated centrifugal detection and the sample quality.

[0112] Specifically, the judgment logic for triggering inertia deviation compensation includes:

[0113] Monitor the sharpness of the stratified edge in real time to determine the compensation trigger condition for inertia deviation compensation;

[0114] Monitor the torque change to judge the authenticity of the compensation trigger condition;

[0115] If the compensation trigger condition is true, adjust the rotational speed in segments to compensate for the inertia deviation;

[0116] Identify the resonance state during the process of adjusting the rotational speed in segments to execute the avoidance mechanism;

[0117] Continuously monitor the change of the sharpness of the stratified edge to verify the compensation effect.

[0118] During the centrifugation of the PRP sample, if the distribution is uneven, it will cause abnormal stratification state, which is reflected in the sharpness of the stratified edge. By continuously monitoring the sharpness of the stratified edge, the signal of abnormal sample distribution can be captured in time, so as to provide a basis for triggering inertia deviation compensation. Among them, a sharpness threshold needs to be set, such as 85%. When the sharpness of the stratified edge is greater than the sharpness threshold, it is initially determined that the compensation trigger condition is met, so as to monitor the sample stratification state in real time and accurately, discover potential inertia deviation problems in time, provide a reliable trigger signal for subsequent compensation operations, and ensure the stable operation of the simulated centrifugation detection and the sample quality.

[0119] The abnormality of the sharpness of the stratified edge is not entirely caused by inertia deviation. There are other interference factors. By simulating the torque change for comprehensive judgment, the inertia deviation caused by uneven sample distribution will lead to abnormal torque. Combining the judgment of torque change can more accurately determine whether there is really inertia deviation and improve the accuracy of the compensation operation. If it is known through the above judgment that the torque change is greater than the change threshold, it is determined that the torque has an abnormal change, that is, when the sharpness of the stratified edge is greater than the sharpness threshold and the torque change is greater than the change threshold, it indicates that the compensation trigger condition is true, thus improving the accuracy of inertia deviation judgment, avoiding unnecessary compensation operations caused by misjudgment, saving energy and time, and at the same time ensuring the stability of the simulated centrifugation detection operation.

[0120] Furthermore, as Figure 4 shown, the sub-logic of adjusting the rotational speed in segments includes:

[0121] When the compensation trigger condition is true, reduce the rotational speed;

[0122] During the process of reducing the rotational speed, monitor the descending amplitude of the torque change. If the descending amplitude of the torque change is greater than the amplitude threshold, restore the rotational speed to the target rotational speed;

[0123] If the descending amplitude of the torque change is less than or equal to the amplitude threshold, identify the resonance state to execute the avoidance mechanism.

[0124] Inertia deviation will cause a large unbalanced force during the simulated centrifugal detection process, resulting in unstable operation of the centrifugal detection system. Reducing the rotational speed can reduce the impact of the unbalanced force on the operation of the simulated centrifugal detection process. At the same time, it is also convenient to observe the torque change situation, providing a buffer for subsequent adjustment. By changing the frequency of the input voltage of the simulated motor with the aid of the simulated motor, the rotational speed of the simulated motor is reduced, thereby correspondingly reducing the centrifugal rotational speed. During the process of reducing the rotational speed, the rotational speed of the simulated motor is monitored in real time to ensure that the rotational speed steadily drops to the set value, thus effectively reducing the impact of inertia deviation on the operation of the simulated centrifugal detection, ensuring the stable operation of the centrifugal detection system and the quality of sample simulation analysis, and at the same time providing a basis for further adjustment according to the torque change in the future.

[0125] During the process of reducing the simulated centrifugal rotational speed, the decreasing amplitude of the torque change is monitored in real time. If the torque decreasing amplitude is large, it indicates that reducing the rotational speed has effectively improved the inertia deviation problem. At this time, restoring the rotational speed to the target rotational speed can ensure the working efficiency of the simulated centrifugal detection and avoid affecting the production progress due to long-term low-speed operation. The decreasing amplitude of the torque change is calculated in real time, that is, the difference between the current torque and the torque before the rotational speed reduction is divided by the time taken for the rotational speed reduction, and an amplitude threshold is configured. When the calculated decreasing amplitude of the torque change is greater than this amplitude threshold, the frequency of the input voltage of the simulated motor is gradually increased with the aid of the simulated motor to restore the rotational speed to the target rotational speed, thereby solving the inertia deviation problem while ensuring the working efficiency of the simulated centrifugal detection and improving the stable operation and service life of the centrifugal detection system.

[0126] When the decreasing amplitude of the torque change is small or not obvious, it indicates the existence of resonance. Resonance will cause severe vibration of the centrifugal detection system, further damaging the centrifugal detection system. Timely identification and avoidance of the resonance state can prevent the centrifugal detection system from being damaged due to resonance and ensure the safe operation of the centrifugal detection system. The vibration signal is transformed from the time domain to the frequency domain through Fourier transform, and the frequency components of the vibration signal are analyzed. When it is found that the vibration frequency is close to the natural frequency of the simulated centrifugal detection system and the torque change appears abnormal, it is determined that there is a resonance state. Once the resonance state is identified, the rotational speed of the simulated motor is quickly adjusted to avoid the resonance frequency by quickly increasing or decreasing the speed, such as increasing or decreasing the speed by a certain proportion, such as 5% to 10%, within a very short time, and then subsequent verification and adaptive adjustment are carried out according to data such as torque change and stratification edge sharpness, thereby effectively preventing the centrifugal detection system from being damaged due to resonance, ensuring the stable operation and service life of the centrifugal detection system, and at the same time improving the effect of inertia deviation compensation.

Claims

1. A centrifugal detection system, characterized in that, Including: A detection module and a control module; The detection module is used to obtain spectral data, image data, and pressure data, extract spectral features, stratification edge sharpness, and pressure change features, perform feature fusion on the spectral features, stratification edge sharpness, and pressure change features through an attention mechanism to obtain a centrifugal fusion feature, and analyze the centrifugal fusion feature to obtain a centrifugal state score; The control module is used to monitor the vibration amplitude and torque change during the simulated centrifugal detection process, and combine the centrifugal state score to judge whether the centrifugal state is abnormal. If the centrifugal state is abnormal, the centrifugal parameters are adjusted, and it is judged whether to trigger sample distribution adjustment and inertia deviation compensation.

2. The centrifugal detection system according to claim 1, wherein The acquisition logic of the spectral data includes: Integrate a dual-wavelength laser and trigger the dual-wavelength laser alternately in time division; Start the acquisition of spectral data when the target sample is in the detection window; Parse the spectral features of the dual-wavelength laser channels in the spectral data respectively.

3. A centrifugal detection system according to claim 2, characterized in that, Obtain image data through an image sensor and pressure data through a pressure sensor. The extraction logic of the stratification edge sharpness includes: Identify the edge pixels of the image data through an edge algorithm, and segment the image data into multiple stratification regions according to the edge pixels of the image data; Configure a local window and calculate the gradient amplitude and gradient direction of the edge pixels of each stratification region within the local window; Comprehensively analyze the gradient amplitude, gradient direction, and local contrast of the edge pixels to obtain the stratification edge sharpness.

4. A centrifugal detection system according to claim 3, wherein, The segmentation sub-logic of the stratification region includes: Generate an adjacency graph based on the edge pixels of the image data. Each node represents an edge pixel, and the edge represents the connection between adjacent edge pixels; Calculate the similarity between nodes and configure a similarity threshold; Cluster the nodes with a similarity greater than the similarity threshold into the same stratification region through a clustering algorithm.

5. An in vitro diagnostic system according to claim 4, wherein The logic of the feature fusion includes: Construct a spatio-temporal matrix of spectral features, stratification edge sharpness, and pressure features, and calculate the spatial attention weight according to the spatio-temporal matrix; Calculate the temporal attention weight according to the spatio-temporal matrix; Perform a Hadamard product on the spatial attention weight and the temporal attention weight to obtain a centrifugal fusion feature.

6. The centrifugal detection system according to claim 5, wherein The judgment logic of whether the centrifugal state is abnormal includes: Determine the physical abnormal state according to the vibration amplitude and torque change during the simulated centrifugal detection process; Configure a score threshold and compare the centrifugal state score with the score threshold to obtain a biological abnormal state; Comprehensively judge whether the centrifugal state is abnormal according to the physical abnormal state and the biological abnormal state.

7. The centrifugal detection system according to claim 6, characterized in that, The judgment logic of triggering sample distribution adjustment includes: Monitor the pressure data in real time to determine the adjustment trigger condition for sample distribution adjustment; Generate an annular pressure gradient map based on the pressure data, mark the pressure distribution area, the pressure distribution area includes a pressure sudden increase area and a pressure sudden drop area, and judge the imbalance direction; Determine the air flow type according to the pressure distribution area and determine the air flow intensity according to the imbalance direction; Update the annular pressure gradient map in real time to verify the adjustment effect.

8. The centrifugal detection system according to claim 7, wherein The imbalance direction is judged according to the distribution mode of the pressure sudden increase area and the pressure sudden drop area. If the pressure sudden increase area and the pressure sudden drop area are symmetrically distributed, it is judged as static imbalance; If the pressure sudden increase area and the pressure sudden drop area are asymmetrically distributed, it is judged as dynamic imbalance.

9. A centrifugal detection system according to claim 8, characterized in that, The judgment logic for triggering inertia deviation compensation includes: Real-time monitoring of the sharpness of the stratified edge to determine the compensation trigger condition for inertia deviation compensation; Monitoring the torque change to judge the authenticity of the compensation trigger condition; If the compensation trigger condition is true, adjust the rotational speed in segments to compensate for the inertia deviation; Identify the resonance state during the process of adjusting the rotational speed in segments to execute the avoidance mechanism; Continuously monitor the change of the sharpness of the stratified edge to verify the compensation effect.

10. A centrifugal detection system according to claim 9, characterized in that, The sub-logic for adjusting the rotational speed in segments includes: When the compensation trigger condition is true, reduce the rotational speed; During the process of reducing the rotational speed, monitor the decreasing amplitude of the torque change. If the decreasing amplitude of the torque change is greater than the amplitude threshold, restore the rotational speed to the target rotational speed; If the decreasing amplitude of the torque change is less than or equal to the amplitude threshold, identify the resonance state to execute the avoidance mechanism.

Images