Intelligent fault early warning and decision method and system of medical centrifuge

Abstract

The application discloses a kind of intelligent fault early warning and decision method and system of medical centrifuge, it is related to operating condition monitoring and intelligent control technical field;The method comprises the following steps: collecting the vibration signal, current signal and rotational speed signal of medical centrifuge acceleration stage, equal-angle resampling is carried out and acceleration complex response quantity is constructed;The background separation is carried out to acceleration complex response quantity, and target narrow speed domain is determined;Double-pulse fitting model is constructed, and double-pulse parameter and fitting credibility are obtained by particle swarm algorithm;Based on double-pulse parameter, extract hysteresis release feature, combined with historical health operation baseline to obtain comprehensive early warning intensity;According to comprehensive early warning intensity, determine target control action and execute.The application can improve the identification and decision accuracy of single-hanging-basket hysteresis release risk.

Description

Technical Field

[0001] This invention relates to the field of operational status monitoring and intelligent control technology, and in particular to an intelligent fault early warning and decision-making method and system for a medical centrifuge. Background Technology

[0002] Medical centrifuges are widely used in blood component separation, clinical laboratory sample pretreatment, and biological sample stratification. The basket-type structure is frequently adopted due to its adaptability to various sample containers and its ability to maintain a relatively stable separation posture after operation. During the acceleration process of such equipment, each basket typically needs to swing out synchronously and gradually as the rotational speed increases, causing continuous changes in the equipment's vibration response and drive load. When a basket experiences slight obstruction, residue buildup, decreased lubrication, or uneven wear at its hinge point, that basket may swing out later than the others, and then release itself as the rotational speed continues to increase. This results in successive abnormal responses within a narrow speed range. These abnormalities are short-lived, occur within a narrow range, and may weaken after reaching a steady state, easily remaining dormant for a long time before any significant equipment failure occurs. Therefore, how to promptly identify abnormalities caused by delayed release of a single basket during the acceleration phase and make appropriate control decisions accordingly has become a crucial technical problem that needs to be solved in improving the operational safety and sample processing stability of basket-type medical centrifuges.

[0003] Existing medical centrifuges mostly employ protection methods based on vibration exceeding limits, asymmetrical loading, or imbalance causing shutdown. While these methods can be effective against significant unbalanced loading or persistent imbalance, they often lack targeted identification and grading mechanisms for short-term anomalies that occur only within a local speed range during the acceleration phase and then quickly subside. On the one hand, traditional methods typically focus on whether the anomaly reaches the shutdown threshold, making it difficult to distinguish between ordinary balancing fluctuations, transient disturbances in the liquid surface, and delayed release of a single basket. On the other hand, existing methods do not pay sufficient attention to the temporal changes before and after the basket swing-out process, making it difficult to extract specific information reflecting early mechanical risks of a single basket from vibration and load changes. This results in the continued existence of false alarms, missed warnings, or simplistic response actions, which in turn affect the continuous operating efficiency of the equipment, the quality of sample processing, and the accuracy of subsequent maintenance decisions. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies, such as inaccurate identification of short-term anomalies in the narrow speed range caused by the delayed release of a single basket during the acceleration phase of a swing-basket medical centrifuge, difficulty in distinguishing between mechanical hysteresis risks and ordinary balancing disturbances, and insufficient targeted early warning and control decisions. Therefore, this invention proposes an intelligent fault early warning and decision-making method and system for medical centrifuges.

[0005] The technical solution of this invention: an intelligent fault early warning and decision-making method for a medical centrifuge, comprising: Vibration signals, current signals, and rotation speed signals of a medical centrifuge during the acceleration phase are collected. Based on the rotation speed signal, equal-angle resampling is performed, and a composite acceleration response is constructed. Background separation is performed on the acceleration composite response to obtain the residual response, and the target narrow velocity range is determined based on the residual response; A bipulse fitting model is constructed based on the residual response within the target narrow velocity domain. The bipulse fitting model is then optimized using a particle swarm optimization algorithm to obtain the bipulse parameters and the bipulse fitting reliability. Based on the dual-pulse parameters, delayed release features are extracted and an evidence vector is constructed. The comprehensive early warning intensity is obtained by combining the abnormal distance between the evidence vector and the historical healthy operation baseline. The delayed release features include hysteresis separation degree, release steepening ratio and post-term residual retention ratio. Based on the comprehensive early warning intensity, the target control action is determined, and the medical centrifuge is controlled to execute the target control action.

[0006] Preferably, vibration signals, current signals, and rotational speed signals of the medical centrifuge are collected during the acceleration phase. Based on the rotational speed signal, equal-angle resampling is performed, and a composite acceleration response is constructed, including: The cumulative rotation angle is calculated based on the rotation speed signal, and the vibration signal, current signal and rotation speed signal are resampled at equal angular intervals to obtain the resampled vibration signal, resampled current signal and angular velocity sequence in the angular velocity domain. Calculate the absolute difference between the resampled vibration signal and its corresponding median, and normalize it based on the median absolute deviation of the resampled vibration signal to obtain the vibration normalized component. Calculate the absolute difference between the resampled current signal and its corresponding median, and normalize it based on the median absolute deviation of the resampled current signal to obtain the normalized current component. The normalized vibration component is added to the normalized current component to obtain the acceleration composite response.

[0007] Preferably, background separation is performed on the acceleration composite response to obtain the residual response, and the target narrow velocity range is determined based on the residual response, including: The background envelope of the acceleration composite response is separated to obtain the residual response; In the residual response, multiple candidate pulse intervals are divided according to the extreme points, and the energy, center velocity, and width of each candidate pulse interval are calculated. The candidate pulse interval with the highest energy is selected as the first candidate pulse interval; Select the candidate pulse interval with higher energy from the candidate pulse intervals that are adjacent to the first candidate pulse interval on the velocity axis as the second candidate pulse interval; The first candidate pulse interval, the second candidate pulse interval, and the valley region between them are merged to form the target narrow velocity domain.

[0008] Preferably, the double-pulse fitting model is optimized using a particle swarm optimization algorithm, including: The dual-pulse parameters include the amplitude, center velocity, and width of the first pulse, and the amplitude, center velocity, and width of the second pulse; Construct an optimization objective function that includes a fitting error term and a structural constraint term, wherein the fitting error term is used to characterize the difference between the double-pulse fitting model and the residual response, and the structural constraint term is used to characterize the proportional relationship between the sum of the first pulse width and the second pulse width and the velocity interval between the two pulse centers; The position of a single particle is defined as a parameter vector consisting of the amplitude, center velocity, and width of the first pulse and the amplitude, center velocity, and width of the second pulse. The particle swarm optimization algorithm is used to minimize the objective function, and the optimal parameters are output as double pulse parameters. The reliability of the double pulse fitting is calculated based on the optimized objective function value after optimization.

[0009] Preferably, extracting delayed release features and constructing an evidence vector based on the dual-pulse parameters includes: Hysteresis separation is obtained based on the relationship between the interval between the center velocities of the first pulse and the second pulse and the sum of the widths of the first pulse and the second pulse. The release steepening ratio is obtained based on the ratio of the amplitude to the width of the second pulse relative to the ratio of the amplitude to the width of the first pulse. The residual retention ratio of the later stage is obtained based on the residual response in the velocity domain after the center velocity of the second pulse plus twice the width of the second pulse. The hysteresis separation degree, the release steepening ratio, and the post-residual retention ratio are combined to form an evidence vector.

[0010] Preferably, the comprehensive early warning intensity is obtained by combining the abnormal distance between the evidence vector and the historical healthy operating baseline, including: The historical healthy operating baseline includes the mean vector and covariance matrix under the same rotor type and similar loading conditions; Calculate the Mahalanobis anomaly distance between the evidence vector and the historical healthy operating baseline, and obtain the delayed release hazard probability based on the Mahalanobis anomaly distance; The overall warning intensity is obtained by multiplying the delayed release danger probability by the double pulse fitting confidence level.

[0011] Preferably, determining the target control action based on the comprehensive early warning intensity includes: The target control actions include continued acceleration, speed maintenance verification, and controlled shutdown; Based on the comprehensive early warning intensity and the residual retention ratio in the later stage, calculate the first utility value corresponding to continued acceleration, the second utility value corresponding to speed maintenance verification, and the third utility value corresponding to controlled shutdown. Compare the first utility value, the second utility value, and the third utility value to determine the target control action.

[0012] Preferably, controlling the medical centrifuge to perform the target control action includes: If the target control action is to continue accelerating, then maintain the current acceleration trajectory; If the target control action is speed-hold verification, the verification rotation speed is set to the center speed of the second pulse, and the verification duration is determined based on the sum of the first pulse width and the second pulse width and the current rate of change of angular velocity. The verification comprehensive warning intensity is recalculated during the verification phase. When the verification comprehensive warning intensity is lower than the original comprehensive warning intensity, the speed increase is resumed. When the verification comprehensive warning intensity is not lower than the original comprehensive warning intensity, a controlled shutdown is executed and a hoisting point check decision is output. If the target control action is controlled shutdown, then deceleration shutdown is executed, and a hysteresis release risk warning and basket hinge point inspection decision are output.

[0013] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: This invention addresses the need for identifying short-term anomalies in the narrow velocity domain corresponding to delayed release of a single basket during the acceleration phase of a pendulum-type medical centrifuge. It establishes a complete analytical chain from acceleration signal acquisition, angular velocity domain resampling, background separation, target narrow velocity domain extraction, to dual-pulse model inversion. By uniformly quantifying the vibration response and driving current response, and extracting the sequentially occurring dual-pulse anomalies on the velocity axis, and combining the hysteresis separation degree, release steepening ratio, and subsequent residual retention ratio to form an evidence vector, it can more accurately characterize the dynamic evolution of the single basket from delayed swing-out to release-type swing-out. This improves the ability to identify hidden mechanical risks during the acceleration phase and enhances the differentiation between ordinary balancing fluctuations, random disturbances, and delayed release anomalies.

[0014] This invention further constructs a comprehensive early warning intensity and hierarchical control decision-making mechanism based on anomaly identification. By quantifying the degree of abnormal deviation between the evidence vector and the historical healthy operating baseline, and combining it with the credibility of double-pulse fitting to form a comprehensive early warning result, the target control action is determined between continued acceleration, speed maintenance verification, and controlled shutdown. This enables medical centrifuges to adopt more suitable operating procedures for different risk levels. It can not only provide more targeted early warnings and basket hinge point inspection decisions when the anomaly is still in its early stages, but also take into account equipment operating efficiency, sample processing stability, and timeliness of fault handling, thereby improving the overall operational safety and the accuracy of subsequent maintenance judgments. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating an intelligent fault warning and decision-making method for a medical centrifuge according to an embodiment of the present invention. Figure 2 This is a functional block diagram of an intelligent fault early warning and decision-making system for a medical centrifuge, provided as an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] Example 1, as Figure 1 As shown, this embodiment provides an intelligent fault early warning and decision-making method for medical centrifuges, specifically including: Vibration signals, current signals, and rotation speed signals of a medical centrifuge during the acceleration phase are collected. Based on the rotation speed signal, equal-angle resampling is performed, and a composite acceleration response is constructed. Background separation is performed on the acceleration composite response to obtain the residual response, and the target narrow velocity range is determined based on the residual response; A bipulse fitting model is constructed based on the residual response within the target narrow velocity domain. The bipulse fitting model is then optimized using a particle swarm optimization algorithm to obtain the bipulse parameters and the bipulse fitting reliability. Based on the dual-pulse parameters, delayed release features are extracted and an evidence vector is constructed. The comprehensive early warning intensity is obtained by combining the abnormal distance between the evidence vector and the historical healthy operation baseline. The delayed release features include hysteresis separation degree, release steepening ratio and post-term residual retention ratio. Based on the comprehensive early warning intensity, the target control action is determined, and the medical centrifuge is controlled to execute the target control action.

[0018] In an embodiment of the present invention, vibration signals, current signals, and rotational speed signals of a medical centrifuge are collected during the acceleration phase. Based on the rotational speed signals, equal-angle resampling is performed, and a composite acceleration response is constructed, including: An accelerometer is radially installed on the casing of the medical centrifuge to collect radial vibration signals. A current acquisition device is installed in the current circuit of the centrifuge drive motor to collect drive current signals. A spindle speed detection device collects spindle speed signals. The sampling frequency of all signals is set to meet the signal acquisition requirements during the acceleration process, and the acquisition range covers the entire acceleration stage of the centrifuge from startup to the preset target speed. The cumulative rotation angle is calculated based on the collected spindle speed signals. The formula for calculating the cumulative rotation angle is as follows: ; in This represents the cumulative rotation angle at time t. express The spindle angular velocity at time t represents the current time. The integration lower limit of 0 corresponds to the start time of the centrifuge. Using the cumulative rotation angle as the resampling benchmark can eliminate the influence of time scale inhomogeneity during the acceleration process, so that subsequent analysis can directly correspond to the rotation angle of the spindle and more accurately capture abnormal responses related to rotation speed.

[0019] The vibration signal, current signal, and rotation speed signal are resampled at equal angular intervals. The equal angular intervals that meet the signal resolution requirements are set. Based on the calculated cumulative rotation angle sequence, the vibration value, current value, and angular velocity value corresponding to each equal angular interval are obtained by linear interpolation. The resampled vibration signal, resampled current signal, and angular velocity sequence in the angular velocity domain are obtained. The linear interpolation has low computational complexity and good real-time performance, which can meet the needs of online monitoring during the centrifuge speed-up process. At the same time, it has sufficient interpolation accuracy for continuously changing signals.

[0020] Calculate the median and median absolute deviation of the resampled vibration signal. Calculate the absolute difference between each sampling point in the resampled vibration signal and its median. Divide this absolute difference by the sum of the median absolute deviation and the minimum value of the resampled vibration signal to obtain the vibration normalized component. The formula for calculating the vibration normalized component is as follows: ; in This represents the normalized vibration component at the k-th sampling point. This represents the resampled vibration signal value at the k-th sampling point. This represents the median of the resampled vibration signal. This represents the median absolute deviation of the resampled vibration signal. To prevent extremely small quantities where the denominator is zero, normalization using the median and median absolute deviation effectively resists the influence of occasional spike interference during acceleration. Compared to the mean and standard deviation, it has stronger robustness and can more accurately reflect the background level of the signal. Using the same calculation method as the vibration normalization component, the median and median absolute deviation of the resampled current signal are calculated. The absolute difference between each sampling point in the resampled current signal and its median is calculated. This absolute difference is divided by the sum of the median absolute deviation and the extremely small quantity of the resampled current signal to obtain the current normalization component. The formula for calculating the current normalization component is: ; in This represents the normalized component of the current at the k-th sampling point. This represents the resampled current signal value at the k-th sampling point. This represents the median of the resampled current signal. This represents the median absolute deviation of the resampled current signal. The value of is the same as that used in vibration normalization calculation. Normalizing the current signal can unify the vibration signal and the current signal to the same dimension, eliminate the influence of the difference in amplitude of different signals on subsequent analysis, and make their contributions to the abnormal response comparable.

[0021] The normalized vibration component and the normalized current component at the same sampling point are added together to obtain the acceleration composite response. The formula for calculating the acceleration composite response is as follows: ; in This represents the acceleration composite response at the kth sampling point. The vibration signal mainly reflects the vibration response of the mechanical structure, while the current signal mainly reflects the load response of the motor. Both represent the operating state of the centrifuge from different dimensions. Integrating the two can simultaneously capture abnormal mechanical vibration and abnormal load fluctuation, improve the sensitivity of identifying the delayed release process of a single basket, and avoid missing detection by a single signal.

[0022] In this embodiment, the signal sampling frequency is preferably set to no less than twenty times the main frequency corresponding to the centrifuge's target maximum speed, to ensure that transient vibration changes and current fluctuations during the acceleration phase can be effectively captured. When the centrifuge's target maximum speed is high or the acceleration slope is large, the sampling frequency can be further increased. The equal angle interval is preferably set to an angle interval that obtains at least thirty resampling points per revolution, to balance the speed resolution required for dual-pulse anomaly identification and the real-time performance of online calculation. The minimum value is preferably taken as one-thousandth to one-hundredth of the corresponding median absolute deviation, or as a positive value not less than the system's numerical calculation resolution, to avoid numerical divergence caused by an excessively small denominator, while ensuring that the normalization result is not significantly disturbed.

[0023] It should be noted that the acceleration phase of a medical centrifuge refers to the continuous increase in speed from startup until the target speed is reached. During this phase, the rotor angular velocity, drive load, and structural response are all in a dynamic state, making it a crucial period for identifying early signs of abnormality. Vibration signals are time-series signals acquired by vibration sensors located on the casing, support components, or adjacent structures, characterizing the strength and trend of mechanical vibration during the acceleration process. These signals reflect operating conditions such as rotor imbalance, basket swaying, structural impact, and localized jamming. Current signals are time-series signals formed by the change in drive current in the power supply circuit of the drive motor over time, characterizing the motor output load, torque demand, and load fluctuations caused by abnormal disturbances. Speed ​​signals are signals used to characterize the spindle unit speed. The time-series signal of rotation speed within a time period corresponds to the centrifuge's operating speed level at each moment and provides a benchmark for subsequent speed domain analysis; equal-angle resampling means that instead of analyzing the original sampled data at fixed time intervals, the same angle rotated by the spindle is used as the uniform sampling interval, mapping the original vibration signal, current signal, and rotation speed signal onto a sequence with consistent angle progression, thereby reducing the impact of uneven time scales on anomaly identification during the acceleration process; the acceleration composite response quantity refers to the comprehensive characterization quantity obtained by normalizing and combining the resampled vibration-related information and current-related information according to a unified rule, used to simultaneously reflect mechanical response anomalies and drive load anomalies, so that hysteresis release symptoms that are not easily manifested in a single physical channel can be highlighted in the unified quantitative results.

[0024] It should be noted that single-basket delayed release refers to a situation in which, during the acceleration process of a swing-basket medical centrifuge, multiple baskets should gradually swing outward from their drooping state and approach their working posture under the action of centrifugal force. However, due to factors such as minor jamming at the hinge, increased contact resistance, local residue adhesion, slight uneven wear, or changes in lubrication, the swinging action of one basket is delayed relative to the other baskets. As the spindle speed continues to increase, the centrifugal torque further increases and eventually overcomes the additional resistance, the basket will release and swing out in a short period of time, thus forming an abnormal process of initial delay followed by sudden swinging out. This phenomenon essentially reflects that a single basket failed to maintain synchronous posture changes with the other baskets during the acceleration phase, causing instantaneous mass distribution imbalance and sudden load response changes in the centrifuge's rotating components within a specific speed range. This may be further superimposed by the sample liquid level redistribution effect, resulting in local abnormal changes with a sequential order in the vibration response and drive current response.

[0025] In an embodiment of the present invention, background separation is performed on the acceleration composite response to obtain the residual response, and a target narrow velocity range is determined based on the residual response, including: The aforementioned acceleration composite response is subjected to background envelope separation. The background envelope of the acceleration composite response is calculated using the moving median filtering method. The formula for calculating the background envelope is as follows: ; in This represents the background envelope value of the k-th sampling point. This represents the moving median filtering operation. This represents the acceleration composite response at the k-th sampling point. This represents the window length of the sliding median filter. The sliding window length is determined based on the total number of sampling points in the current acceleration segment, and is usually between 1 / 4 and 1 / 10 of the total number of sampling points in the current acceleration segment, in order to balance the background suppression capability and the anomaly retention capability. Using the sliding median filter for background separation can effectively suppress slowly changing baseline drift and random noise interference during acceleration. Compared with the sliding mean filter, it has stronger robustness against outliers and can more accurately extract local sudden abnormal responses.

[0026] Calculate the difference between the acceleration composite response and the background envelope, and take the non-negative value of the difference as the residual response. The formula for calculating the residual response is: ; in The residual response value at the k-th sampling point is represented by the value of the residual response. Only non-negative values ​​are retained to filter out the negative bias caused by background fluctuations, so that the residual response only reflects positive anomalies above the background level, avoiding interference from irrelevant fluctuations on subsequent pulse identification. The local maximum point is defined as the point whose value is greater than the values ​​of the two adjacent sampling points on the left and right, and the local minimum point is defined as the point whose value is less than the values ​​of the two adjacent sampling points on the left and right. All local maximum points and the first local minimum point on the left and right sides of each local maximum point are searched in the residual response. The interval between each local maximum point and the first local minimum point on the left and right sides is defined as a candidate pulse interval, thereby dividing the residual response into multiple non-overlapping candidate pulse intervals.

[0027] The energy, center velocity, and width of each candidate pulse interval are calculated sequentially. The formula for calculating the energy of a candidate pulse interval is as follows: ; in This represents the energy of the q-th candidate pulse interval. Let represent the set of sampling points corresponding to the q-th candidate pulse interval. Using the sum of squared residuals to calculate pulse energy amplifies the differences in anomalous responses, making higher-energy anomalous pulses easier to identify. The formula for calculating the center velocity of the candidate pulse interval is: ; in This represents the center velocity of the q-th candidate pulse interval. This represents the angular velocity value corresponding to the k-th sampling point. To prevent extremely small quantities where the denominator is zero, the energy-weighted average calculation of the center velocity more accurately reflects the concentration of pulse energy and better reflects the actual occurrence speed of abnormal responses compared to the arithmetic mean; the formula for calculating the candidate pulse interval width is: ; in The width of the q-th candidate pulse interval is represented by the pulse width calculated using energy-weighted standard deviation, which can accurately reflect the dispersion of the pulse on the velocity axis and quantify the duration of the abnormal response.

[0028] Compare the energies of all candidate pulse intervals and select the one with the highest energy as the first candidate pulse interval. Traverse all candidate pulse intervals and select those that are located after and adjacent to the first candidate pulse interval on the velocity axis. Select the one with the highest energy from the selected candidate pulse intervals as the second candidate pulse interval. Merge the set of sampling points corresponding to the first candidate pulse interval, the set of sampling points corresponding to the second candidate pulse interval, and all the set of sampling points between the two candidate pulse intervals to obtain the set of sampling points corresponding to the target narrow velocity domain. Determine the velocity domain corresponding to this set of sampling points. Merging the two candidate pulse intervals and the intermediate valley can completely cover the continuous double-pulse abnormal response generated by the single basket hysteresis release process, providing a complete analysis range for subsequent double-pulse model inversion. If no two consecutive candidate pulse intervals that meet the requirements are detected, it is determined that there is no narrow velocity domain double-pulse abnormality, and the acceleration action is directly executed.

[0029] In this embodiment, the failure to detect two consecutive candidate pulse intervals that meet the requirements means that there are no two candidate pulse intervals in the residual response of the current acceleration stage that satisfy the consecutive adjacent relationship and both have effective pulse energy. If only a single candidate pulse interval is detected, two candidate pulse intervals have obvious overlap and cannot be distinguished as consecutive, or the energy of the candidate pulse interval is lower than the background disturbance level, it is determined that a narrow velocity domain double pulse anomaly corresponding to single basket hysteresis release has not been formed. At this time, continuing the acceleration action only means that no special verification or shutdown decision for single basket hysteresis release is triggered, and does not exclude the medical centrifuge from performing independent alarm or protection control for other types of abnormal operating conditions according to its conventional protection logic. The narrow-velocity double-pulse anomaly corresponding to the delayed release of a single basket refers to a situation in a swing-basket medical centrifuge where, during the acceleration process, a basket fails to swing out synchronously with the others due to excessive resistance at the hinge. Initially, a local imbalance response caused by the delayed swing-out occurs at a lower speed range. Subsequently, as the rotational speed continues to increase and the centrifugal torque increases and overcomes the resistance, the basket releases and swings out again within a short speed span. This is further superimposed with additional disturbances caused by the redistribution of liquid within the container, resulting in two consecutive local abnormal responses occurring within a narrow, adjacent speed range on the speed axis. The first pulse mainly characterizes the transient imbalance accumulation caused by the delayed swing-out of the single basket, while the second pulse mainly characterizes the sudden enhanced response caused by the basket's release and liquid redistribution. Together, they constitute a double-pulse anomaly pattern that occurs only within a specific narrow speed range. This anomaly can be distinguished from ordinary balancing fluctuations or random disturbances and is an important sign for identifying early mechanical hysteresis risks in single baskets.

[0030] It should be noted that the residual response refers to the local anomalous response retained after removing the background trend from the acceleration composite response. It mainly corresponds to sudden changes above the background level and is an important basis for identifying double-pulse anomalies. The candidate pulse interval refers to the velocity interval enclosed by the local high point as the core and the adjacent boundary positions on both sides. This interval is used to carry a relatively complete local anomalous pulse response. The energy refers to the quantitative result of accumulating the residual response intensity within the candidate pulse interval. It reflects the overall strength of the anomalous response within the candidate pulse interval. The higher the energy, the more prominent the anomaly within the interval. The center velocity refers to the representative velocity position where the anomalous response is mainly concentrated within the candidate pulse interval. It is used to characterize the main occurrence velocity of the pulse anomaly during the acceleration process. The width refers to the extent of the anomalous response along the velocity direction within the candidate pulse interval. It is used to reflect the velocity span and the degree of dispersion of the anomaly.

[0031] It should be noted that the first candidate pulse interval refers to the interval with the highest energy among all candidate pulse intervals, which usually corresponds to the most significant anomalous response; the velocity axis refers to the velocity coordinate sequence established in order of centrifuge speed change, used to characterize the sequential positional relationship of different anomalous responses during the acceleration process; the second candidate pulse interval refers to the candidate pulse interval that is adjacent to the first candidate pulse interval in velocity sequence and has higher energy, which is used together with the first candidate pulse interval to characterize the double-pulse anomalies that occur successively during the hysteresis release process; the valley region refers to the transition interval between the first and second candidate pulse intervals, where the residual response is relatively reduced, which characterizes the interval and connection state between the two anomalous responses; the target narrow velocity domain refers to the continuous velocity interval composed of the first candidate pulse interval, the second candidate pulse interval, and the valley region between them, which is used to completely cover the range of double-pulse anomalies corresponding to the single basket hysteresis release and serves as a dedicated analysis interval for subsequent double-pulse model fitting and feature extraction.

[0032] In an embodiment of the present invention, a bipulse fitting model is constructed based on the residual response within the target narrow velocity domain. The bipulse fitting model is then optimized using a particle swarm optimization algorithm to obtain the bipulse parameters and the bipulse fitting reliability, including: A two-pulse fitting model is constructed based on the residual response within a defined target narrow velocity domain. The two-pulse fitting model used is a double Gaussian pulse model, and the calculation formula for the double Gaussian pulse model is as follows: ; in The target's angular velocity within a narrow velocity range is... The model fit value at that time, This indicates the amplitude of the first pulse. This indicates the amplitude of the second pulse. Indicates the center velocity of the first pulse. Indicates the center velocity of the second pulse. Indicates the width of the first pulse. The width of the second pulse is indicated by the fact that the center velocity of the first pulse is less than that of the second pulse. The use of the double Gaussian pulse model can accurately match the narrow velocity domain double pulse imbalance phenomenon caused by the delayed release of the single basket, and fit the physical distribution of the abnormal response.

[0033] Construct an optimization objective function that includes a fitting error term and a structural constraint term. The formula for calculating the optimization objective function is as follows: ; in This indicates the value of the objective function to be optimized. This represents the parameter vector to be optimized. This represents the angular velocity value corresponding to the k-th sampling point within the target's narrow velocity domain. Indicates the target narrow velocity range, This represents the residual response value at the k-th sampling point within the target narrow velocity domain. This represents the model fit value for the corresponding sampling point. The term represents a very small amount to prevent the denominator from being zero. The fitting error term is the first term in the formula, which is used to quantify the fitting deviation between the double-pulse fitting model and the actual residual response, thereby constraining the model fitting accuracy. The structural constraint term is the second term in the formula, which is used to limit the double pulses to maintain separable structural features on the velocity axis, avoiding excessive overlap between the two pulses and making them indistinguishable, thus matching the physical characteristics of the double pulses generated by the hysteresis release of the single basket.

[0034] In the particle swarm optimization (PSO) algorithm, the position of a single particle is defined as a parameter vector composed of the following parameters in sequence: amplitude of the first pulse, center velocity of the first pulse, width of the first pulse, amplitude of the second pulse, center velocity of the second pulse, and width of the second pulse. During the PSO optimization process, reasonable value range constraints are set for each parameter. Specifically, the amplitude parameter is greater than 0, the center velocity parameter is within the target narrow velocity range, the width parameter is greater than 0, and the center velocity of the first pulse is less than the center velocity of the second pulse. The number of particles is typically 20 to 50, the number of iterations is typically 50 to 200, the inertia weight is typically 0.4 to 0.9, and the learning factor is typically 1.5 to... 2.5 The above parameters can be adjusted according to actual calculation accuracy and real-time requirements. The particle velocity and position are updated according to the iterative rules of the particle swarm optimization algorithm. The particle velocity update is corrected by combining the historical best position and the global best position with the inertial weight. The particle position is iteratively adjusted based on the updated velocity. The optimization objective function is minimized. Through multiple iterative searches, the globally optimal parameters that minimize the optimization objective function are obtained, and these globally optimal parameters are output as the final double pulse parameters. The double pulse fitting reliability is calculated based on the optimal optimization objective function value obtained after optimization. The formula for calculating the double pulse fitting reliability is: ; in Indicates the reliability of the double-pulse fitting. This represents the optimal parameter vector obtained by the particle swarm optimization algorithm. This represents the value of the objective function corresponding to the optimal parameters. Using an exponential function to calculate the fitting confidence level can normalize the value of the objective function to a confidence value between zero and one. The smaller the value of the objective function, the higher the fitting confidence level, which intuitively represents the reliability of the double-pulse model in fitting the actual residual response.

[0035] It should be noted that the dual-pulse parameters refer to a set of model parameters used to jointly characterize the morphology of two consecutive local abnormal responses within a narrow velocity domain of the target. They are used to characterize the distribution of dual-pulse anomalies along the velocity axis during the delayed release of a single basket. The amplitude of the first pulse refers to the peak intensity of the abnormal response corresponding to the first pulse, which reflects the strength of the local imbalance response during the delayed phase. The center velocity of the first pulse refers to the representative velocity position where the energy of the abnormal response of the first pulse is most concentrated along the velocity axis, which characterizes the velocity level at which the delayed anomaly mainly occurs. The width of the first pulse refers to the range of the first pulse along the velocity direction, which characterizes the velocity span and dispersion of the sustained abnormal response. The amplitude of the second pulse refers to the peak intensity of the abnormal response corresponding to the second pulse, which reflects the strength of the sudden enhanced response caused by the release and swing out of the basket and the redistribution of the liquid surface. The center velocity of the second pulse refers to the representative velocity position where the energy of the abnormal response of the second pulse is most concentrated along the velocity axis, which characterizes the velocity level at which the release swing out mainly occurs. The width of the second pulse refers to the range of the second pulse along the velocity direction, which characterizes the velocity span at which the abnormal response continues after release.

[0036] It should be noted that the optimization objective function is a comprehensive discriminant function used to evaluate whether a set of candidate double-pulse parameters can reasonably describe the actual abnormal response. The smaller the value of the objective function, the better the corresponding parameters can simultaneously meet the requirements of fitting accuracy and structural rationality. The fitting error term is a component in the optimization objective function used to measure the degree of difference between the output result of the double-pulse fitting model and the actual residual response. It mainly reflects the degree of fit of the model to the real abnormal shape under the current parameters. The structural constraint term is a component in the optimization objective function used to restrict the relationship between the two pulse shapes. It constrains the ratio of the sum of the widths of the first pulse and the second pulse to the velocity interval between the centers of the two pulses, so that the two pulses maintain a distinguishable sequential structure and avoid excessive overlap of the two abnormalities, thus losing the physical characteristics of the single basket hysteresis release. The double-pulse fitting model is a mathematical expression model that uses two pulse functions to jointly represent the distribution of the abnormal response in the target narrow velocity domain. Its function is to reconstruct the double-pulse abnormal shape corresponding to the single basket hysteresis release in a parameterized form.

[0037] It should be noted that the reliability of the double pulse fitting refers to the quantitative result representing the reliability of the current double pulse model, which is obtained by transforming the value of the optimized objective function after optimization. It is used to reflect the explanatory power and matching degree of the obtained double pulse parameters to the actual abnormal response. The higher the fitting reliability, the more the currently identified double pulse anomaly matches the physical response characteristics corresponding to the single basket hysteresis release.

[0038] In an embodiment of the present invention, delayed release features are extracted based on the dual-pulse parameters and an evidence vector is constructed. The comprehensive early warning intensity is obtained by combining the abnormal distance between the evidence vector and the historical healthy operating baseline, including: Based on the dual-pulse parameters obtained through particle swarm optimization, the hysteresis separation degree, release steepening ratio, and subsequent residual retention ratio are calculated sequentially. First, the hysteresis separation degree is calculated using the following formula: ; in Indicates the degree of hysteresis separation. Indicates the center velocity of the first pulse. Indicates the center velocity of the second pulse. Indicates the width of the first pulse. Indicates the width of the second pulse. The hysteresis separation degree is used to quantify the degree of separation of the two pulses on the velocity axis. This parameter can intuitively reflect the interval characteristics of the two physical processes of single basket hysteresis release and subsequent release swing in the speed dimension. It is the core basis for distinguishing between double pulse anomalies and ordinary disturbances.

[0039] Calculate the release steepening ratio. The formula for calculating the release steepening ratio is: ; in Indicates the release steepening ratio, This indicates the amplitude of the first pulse. The amplitude of the second pulse is indicated by the release steepening ratio, which characterizes the degree of energy increase of the second pulse compared to the first pulse. It can accurately match the violent transient response generated when the single basket overcomes the jamming and releases the swing, and effectively distinguish between mechanical hysteresis release and conventional balancing fluctuations.

[0040] Calculate the residual retention ratio for the latter part of the error. The formula for calculating the residual retention ratio for the latter part of the error is: ; Where Q represents the residual retention ratio in the latter part. This represents the angular velocity value corresponding to the k-th sampling point. The formula represents the residual response value at the kth sampling point. The formula selects the velocity range after adding twice the width of the second pulse to the center velocity of the second pulse as the statistical range. The numerator is the sum of squares of all residual responses in this range, and the denominator is the characteristic value of the total energy of the double pulse. The residual retention ratio in the latter part is used to quantify the degree of decline of the abnormal response after the appearance of the double pulse, which can accurately identify the abnormally rapid decay performance after the hysteresis release of the single basket.

[0041] The calculated hysteresis separation ratio and the residual retention ratio are combined in a fixed order to form an evidence vector, the expression of which is: ; in The evidence vector integrates the spatial separation features, energy abrupt change features, and subsequent attenuation features of the dual pulses. It fully carries all the core discrimination information of the delayed release caused by the micro-clamping of the single basket hinge point, providing standardized feature input for subsequent anomaly distance calculation and early warning intensity assessment.

[0042] It should be noted that hysteresis separation is a quantitative result obtained from the relationship between the interval between the center velocities of two pulses and the sum of the widths of the two pulses. It is used to measure whether the two pulses have a clear sequential separation structure. The larger the hysteresis separation, the more obvious the separation of the two anomalies on the velocity axis, and the more consistent it is with the physical evolution process of single-basket hysteresis release. Release steepness ratio is a quantitative result formed by the ratio of the amplitude to the width of the second pulse to the ratio of the amplitude to the width of the first pulse. It is used to characterize the steepness of the anomaly enhancement during the release swing phase. The larger the release steepness ratio, the stronger the suddenness and concentration of the subsequent pulse compared to the previous pulse. Velocity range refers to the continuous velocity range defined with the velocity axis as the reference. The center velocity of the second pulse plus the two pulses... The velocity domain interval after the second pulse width refers to the continuous velocity region located after the main distribution range of the subsequent pulse. This region is used to observe the residual disturbance state after the end of the double-pulse anomaly. The residual retention ratio is a quantitative result obtained based on the degree of retention of the residual response in the aforementioned subsequent velocity domain interval. It is used to characterize whether the response drops rapidly after the end of the double-pulse anomaly. The smaller the residual retention ratio, the faster the anomaly decays after release, which is more consistent with the transient characteristics corresponding to the single-basket hysteresis release. The evidence vector is a feature set formed by combining the hysteresis separation degree, release steepening ratio, and residual retention ratio in a fixed order. It is used to uniformly express the separation characteristics, sudden enhancement characteristics, and subsequent decay characteristics of the double pulse as a standardized input for subsequent risk judgment and comprehensive early warning intensity calculation.

[0043] The historical healthy operation baseline is obtained by collecting and statistically analyzing multiple normal speed-up operation data under the same rotor type and similar loading conditions. Typically, at least 10 normal speed-up operation data without abnormalities are collected for statistical analysis to ensure the statistical reliability of the baseline. The historical healthy operation baseline includes a mean vector and a covariance matrix. The mean vector is the overall statistical mean of the evidence vector under healthy conditions, and the covariance matrix is ​​used to characterize the degree of linear correlation between the various dimensions of the evidence vector under healthy conditions. Among these, similar loading conditions preferably refer to loading states where at least the main influencing factors remain consistent: the type of sample container used is consistent, the rotor type and the number of baskets are consistent, the symmetrical loading method is consistent, the number of sample loads is consistent or the difference is within the allowable balancing range, and the loading rate of a single sample container is within the same loading level range. To improve the stability of the historical healthy operation baseline, it is preferable to collect normal speed-up operation data under the same rotor, same container type, and same target speed conditions. When the loading conditions change significantly, a new historical healthy operation baseline can be established, or separate historical healthy operation baselines can be saved for different loading conditions.

[0044] Calculate the Mahalanobis anomaly distance between the evidence vector and the historical healthy operating baseline. The formula for calculating the Mahalanobis anomaly distance is: ; Where M represents the Mahalanobis anomaly distance. This represents the currently calculated evidence vector. The mean vector representing the historical baseline of healthy operation. The covariance matrix represents the historical baseline of healthy operation, with superscript... This represents the matrix transpose operation. The inverse matrix represents the inverse of the covariance matrix. Mahalanobis anomaly distance can effectively eliminate the dimensional differences and linear correlations of the features of each dimension of the evidence vector, accurately measure the deviation of the current operating state from the healthy state, and provide a quantitative indicator for fault risk assessment. When the covariance matrix is ​​ill-conditioned, has an excessively large condition number, or is irreversible, the covariance matrix can be regularized before inverting it. The regularization process includes superimposing a preset small positive value on the diagonal of the covariance matrix; or using the generalized inverse matrix of the covariance matrix to replace the inverse matrix for Mahalanobis anomaly distance calculation. Through the above processing, the numerical stability of the anomaly distance calculation under different operating batches and different normal sample numbers can be guaranteed.

[0045] The probability of delayed release danger is calculated based on the Markov anomaly distance. The formula for calculating the probability of delayed release danger is as follows: ; in This indicates the probability of a delayed release hazard. It represents exponential operations with the natural constant as the base. This represents the Mahalanobis anomaly distance. This calculation method normalizes the Mahalanobis anomaly distance to a numerical range of zero to one. The larger the Mahalanobis anomaly distance, the higher the probability of delayed release. It can intuitively quantify the risk level of delayed release caused by micro-jamming at the hinge point of a single basket.

[0046] The comprehensive warning intensity is obtained by multiplying the probability of delayed release danger by the confidence level of the double-pulse fitting. The formula for calculating the comprehensive warning intensity is as follows: ; in Indicates the overall warning intensity. This indicates the probability of a delayed release hazard. The comprehensive early warning intensity represents the reliability of the double pulse fitting. It integrates the degree of abnormal deviation of the operating status with the fitting reliability of the double pulse model, which not only eliminates the risk of misjudgment caused by model fitting failure, but also accurately reflects the true severity of the single basket delayed release fault, providing a stable and objective core quantitative basis for subsequent graded control decisions.

[0047] It should be noted that the probability of delayed release danger refers to the risk quantification result obtained by the deviation of the evidence vector extracted from the current operating cycle from the historical healthy operating baseline. It is used to characterize how likely the current double pulse anomaly is to be caused by the delayed release process due to the micro-jamming of the single basket hinge point. The historical healthy operating baseline corresponds to the statistical characteristics of the normal operating state under the same rotor type and the same loading conditions. The greater the deviation of the evidence vector from the baseline, the less the current operating state is in line with the healthy operating conditions, and the higher the probability of delayed release danger. Therefore, this quantity actually reflects the probability and severity of the current anomaly belonging to the single basket delayed release fault risk. The comprehensive warning intensity refers to the comprehensive quantitative result obtained by combining the probability of delayed release danger with the reliability of double pulse fitting. It not only considers the degree of abnormal deviation of the current operating state from the healthy baseline, but also the reliability of the matching between the identified double pulse pattern and the narrow velocity domain double pulse anomaly corresponding to the single basket delayed release. Therefore, it can simultaneously reflect the risk level of the anomaly itself and the reliability of the anomaly pattern identification result. The higher the comprehensive warning intensity, the more it indicates that the current operating process not only shows the risk characteristics of a significant deviation from the healthy baseline, but also shows a double pulse structure consistent with the physical response of the single basket delayed release, and more stringent verification or shutdown decisions should be triggered.

[0048] In an embodiment of the present invention, determining a target control action based on the comprehensive early warning intensity and controlling the medical centrifuge to execute the target control action includes: The predefined target control actions include three categories: continued acceleration, speed maintenance verification, and controlled shutdown. Based on the comprehensive warning intensity and residual retention ratio obtained from the previous calculations, the utility values ​​corresponding to the three types of control actions are calculated respectively. The first utility value corresponding to continued acceleration is calculated, and the formula for calculating the first utility value is as follows: ; in This represents the first utility value corresponding to continued acceleration. Indicates the overall warning intensity. This represents the residual retention ratio in the later stage. This utility value increases as the overall warning intensity and the residual retention ratio in the later stage decrease. When the overall warning intensity is low and the abnormal response drops rapidly, the utility of continuing to increase the speed is the highest, which meets the control requirements under normal operating conditions.

[0049] Calculate the second utility value corresponding to the rate-holding verification. The formula for calculating the second utility value is as follows: ; in This represents the second utility value corresponding to the sustained-speed verification. This utility value reaches its peak when the overall warning intensity is high but the residual in the later stage remains relatively low. This corresponds to the working condition where there is a suspected anomaly but the anomaly response has quickly fallen back. At this time, the nature of the anomaly can be further confirmed through sustained-speed verification, which can help to identify potential risks while avoiding accidental shutdown.

[0050] Calculate the third utility value corresponding to the controlled shutdown. The formula for calculating the third utility value is as follows: ; in This represents the third utility value corresponding to controlled shutdown. This utility value increases with the increase of the overall warning intensity and the residual retention ratio in the later stage. When the overall warning intensity is high and the abnormal response does not effectively decline, the utility of controlled shutdown is the highest, which can terminate the dangerous operating state in time and protect the safety of equipment and samples.

[0051] By comparing the first, second, and third utility values, the control action corresponding to the largest utility value is selected as the final target control action. The utility maximization decision-making method can quantify the benefits and risks of different control actions, achieve the optimal balance between operating efficiency and equipment safety, and provide graded and precise control strategies for different abnormal conditions.

[0052] It should be noted that "continue to accelerate" refers to a control method where, when the control system determines that the current anomaly risk is low and subsequent disturbances show a rapid decline characteristic, it maintains the predetermined acceleration trajectory to allow the centrifuge to continue running towards the target speed. This action corresponds to a handling strategy that prioritizes normal operation. "Hold speed for verification" refers to a control method where, when the control system determines that there is a certain risk of delayed release but the anomaly has not yet reached the level requiring immediate shutdown, it keeps the centrifuge running continuously for a period of time near a specific verification speed to re-acquire and analyze signals, thereby further distinguishing between occasional disturbances and persistent mechanical risks. This action corresponds to a handling strategy that prioritizes risk review. "Controlled shutdown" refers to a control method where, when the control system determines that the current anomaly risk is high or the anomaly has not effectively declined within the subsequent speed range, it decelerates the centrifuge in an orderly manner according to a preset deceleration method and stops operation. This action corresponds to a handling strategy that prioritizes equipment and sample safety.

[0053] It should be noted that the first utility value refers to the comprehensive evaluation result of the benefits and risks calculated for the continued acceleration action, which is used to characterize the rationality of continuing operation under the current risk conditions. The second utility value refers to the comprehensive evaluation result of the benefits and risks calculated for the sustained speed verification action, which is used to characterize the rationality of further confirming the abnormal nature through additional verification. The third utility value refers to the comprehensive evaluation result of the benefits and risks calculated for the controlled shutdown action, which is used to characterize the rationality of immediately terminating operation to avoid the expansion of the fault.

[0054] The centrifuge control process is executed according to the determined target control action. If the target control action is to continue to increase the speed, the system will continue to run along the preset speed increase trajectory until the target speed is reached and the speed increase process is completed. This control method can ensure the operating efficiency of the centrifuge under normal operating conditions and avoid unnecessary operation interruptions.

[0055] If the target control action is speed-hold verification, the verification rotation speed is set to the center speed of the second pulse, and the verification duration is determined based on the sum of the first pulse width and the second pulse width, as well as the current rate of change of angular velocity. The formula for calculating the verification duration is as follows: ; in Indicates the verification duration. Indicates the width of the first pulse. Indicates the width of the second pulse. This represents the rate of change of instantaneous angular velocity corresponding to the center velocity of the second pulse. To prevent extremely small amounts where the denominator is zero, the verification rotation speed is set to the center speed of the second pulse because this speed corresponds to the location where the single basket release swing occurs, and the abnormal response is most significant at this point, maximizing the sensitivity of the verification. The calculation method of the verification duration can fully cover the time length corresponding to the continuous speed range of the double-pulse abnormality, ensuring that the verification process can fully capture potential continuous abnormalities. During the verification phase, the system maintains stable operation at the verification rotation speed, and at the same time, according to the same signal acquisition and data processing procedure as the acceleration phase, it re-acquires operating data and calculates the comprehensive verification warning intensity. When the comprehensive verification warning intensity is lower than the original comprehensive warning intensity, the abnormality is determined to be an occasional liquid level fluctuation or temporary balancing disturbance, and the system resumes the original acceleration trajectory and continues to operate. When the comprehensive verification warning intensity is not lower than the original comprehensive warning intensity, the abnormality is determined to be a continuous mechanical risk caused by a slight jamming of the single basket hinge point, and the system immediately executes controlled shutdown and outputs a basket hinge point inspection decision.

[0056] If the target control action is controlled shutdown, the system will perform deceleration shutdown according to the preset deceleration curve. The deceleration curve adopts the conventional linear deceleration or piecewise linear deceleration curve of centrifuges to avoid the impact of sudden stop on samples and equipment. At the same time, it outputs a risk warning of delayed release of single basket and a basket hinge point inspection decision, prompting maintenance personnel to check and maintain the hinge structure of each basket one by one, and eliminate early mechanical failure hazards in a timely manner.

[0057] It should be noted that the delayed release risk warning refers to the control system identifying and evaluating the vibration response, current response, and abnormal double-pulse structure during the acceleration process of the medical centrifuge. If the system determines that the current operating state presents a risk of delayed release caused by localized obstruction, adhesion, uneven wear, or changes in lubrication at the single basket hinge, it outputs a risk warning message to maintenance personnel or the upper control unit. This risk warning message indicates that the current anomaly is not a common balancing fluctuation or random disturbance, but rather a potential mechanical fault symptom with a physical indication of delayed release in the single basket. This provides a basis for subsequent actions such as speed-controlled verification, controlled shutdown, or manual maintenance. The basis for this decision is the maintenance instruction issued by the control system to the hinge structure of the basket in a swing-basket medical centrifuge after determining that the risk of delayed release has reached a level requiring further investigation. The purpose is to guide maintenance personnel to conduct targeted inspections of the hinge joints between each basket and the rotor, focusing on whether there are adverse factors affecting synchronous swinging, such as increased rotational resistance, localized jamming, residue adhesion, abnormal wear of contact surfaces, decreased lubrication, or slight deformation. This allows for timely identification of the potential root cause of delayed release in a single basket and the implementation of maintenance measures, thereby preventing such early mechanical risks from further amplifying in subsequent operation.

[0058] Only when the current running result is to continue accelerating and complete the run, or when the speed maintenance verification passes and the acceleration is resumed and completed, is the historical healthy operating baseline adaptively updated to avoid data pollution of the health model under abnormal operating conditions. The mean vector and covariance matrix of the historical healthy operating baseline are updated incrementally, and the update formula is as follows: ; ; in This represents the updated mean vector. Let represent the updated covariance matrix. This represents the mean vector before the update. This represents the covariance matrix before the update. This represents the evidence vector under the current normal operating conditions. This represents the preset update step size, which is typically between 0.01 and 0.1. A smaller step size results in slower baseline updates but higher statistical stability; a larger step size allows the baseline to better adapt to changes in equipment status. Incremental updates, while preserving the statistical characteristics of historical health data, gradually integrate with the machine's operation, forming an individualized health model tailored to the specific rotor and loading conditions of the machine. This continuously reduces the probability of false alarms in subsequent operations and improves decision-making accuracy. Only evidence vectors from completed speed-up operations or those that have passed speed-hold verification and returned to normal operation are included in the health baseline update. Operation cycles that trigger controlled shutdowns, fail speed-hold verification, or have other protective alarms are not included in the health baseline update. The updated mean vector and covariance matrix are used for calculating anomaly distances under subsequent rotor types and loading conditions, thus allowing the historical healthy operating baseline to gradually adapt to long-term changes in the machine's operating status while avoiding contamination of the health model by abnormal samples.

[0059] like Figure 2 The diagram shown is a functional block diagram of an intelligent fault early warning and decision-making system for a medical centrifuge provided in an embodiment of the present invention.

[0060] In this embodiment, the functions of each module / unit are as follows: The data construction module is used to collect vibration signals, current signals and speed signals during the acceleration phase of a medical centrifuge, perform equal-angle resampling based on the speed signals, and construct the acceleration composite response quantity. The interval identification module is used to perform background separation on the acceleration composite response, obtain the residual response, and determine the target narrow velocity range based on the residual response; The parameter inversion module is used to construct a double-pulse fitting model based on the residual response in the target narrow velocity domain, optimize the double-pulse fitting model using the particle swarm optimization algorithm, and obtain the double-pulse parameters and the double-pulse fitting reliability. The early warning assessment module is used to extract hysteresis release features based on the dual-pulse parameters and construct an evidence vector. Combined with the abnormal distance between the evidence vector and the historical healthy operation baseline, the comprehensive early warning intensity is obtained. The hysteresis release features include hysteresis separation degree, release steepening ratio and post-term residual retention ratio. The decision control module is used to determine the target control action based on the comprehensive early warning intensity and control the medical centrifuge to execute the target control action.

[0061] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for intelligent fault early warning and decision-making in a medical centrifuge, characterized in that, include: Vibration signals, current signals, and rotation speed signals of a medical centrifuge during the acceleration phase are collected. Based on the rotation speed signal, equal-angle resampling is performed, and a composite acceleration response is constructed. Background separation is performed on the acceleration composite response to obtain the residual response, and the target narrow velocity range is determined based on the residual response; A bipulse fitting model is constructed based on the residual response within the target narrow velocity domain. The bipulse fitting model is then optimized using a particle swarm optimization algorithm to obtain the bipulse parameters and the bipulse fitting reliability. Based on the dual-pulse parameters, delayed release features are extracted and an evidence vector is constructed. The comprehensive early warning intensity is obtained by combining the abnormal distance between the evidence vector and the historical healthy operation baseline. The delayed release features include hysteresis separation degree, release steepening ratio and post-term residual retention ratio. Based on the comprehensive early warning intensity, the target control action is determined, and the medical centrifuge is controlled to execute the target control action; Vibration signals, current signals, and rotational speed signals of a medical centrifuge during the acceleration phase are collected. Based on the rotational speed signal, equal-angle resampling is performed, and a composite acceleration response is constructed, including: The cumulative rotation angle is calculated based on the rotation speed signal, and the vibration signal, current signal and rotation speed signal are resampled at equal angular intervals to obtain the resampled vibration signal, resampled current signal and angular velocity sequence in the angular velocity domain. Calculate the absolute difference between the resampled vibration signal and its corresponding median, and normalize it based on the median absolute deviation of the resampled vibration signal to obtain the vibration normalized component. Calculate the absolute difference between the resampled current signal and its corresponding median, and normalize it based on the median absolute deviation of the resampled current signal to obtain the normalized current component. The vibration normalized component is added to the current normalized component to obtain the acceleration composite response. Background separation is performed on the acceleration composite response to obtain the residual response, and the target narrow velocity range is determined based on the residual response, including: The background envelope of the acceleration composite response is separated to obtain the residual response; In the residual response, multiple candidate pulse intervals are divided according to the extreme points, and the energy, center velocity, and width of each candidate pulse interval are calculated. The candidate pulse interval with the highest energy is selected as the first candidate pulse interval; Select the candidate pulse interval with higher energy from the candidate pulse intervals that are adjacent to the first candidate pulse interval on the velocity axis as the second candidate pulse interval; The first candidate pulse interval, the second candidate pulse interval, and the valley region between them are merged to form the target narrow velocity domain. The double-pulse fitting model is optimized using a particle swarm optimization algorithm, including: The dual-pulse parameters include the amplitude, center velocity, and width of the first pulse, and the amplitude, center velocity, and width of the second pulse; Construct an optimization objective function that includes a fitting error term and a structural constraint term, wherein the fitting error term is used to characterize the difference between the double-pulse fitting model and the residual response, and the structural constraint term is used to characterize the proportional relationship between the sum of the first pulse width and the second pulse width and the velocity interval between the two pulse centers; The position of a single particle is defined as a parameter vector consisting of the amplitude, center velocity, and width of the first pulse and the amplitude, center velocity, and width of the second pulse. The particle swarm optimization algorithm is used to minimize the objective function, and the optimal parameters are output as double pulse parameters. The reliability of the double pulse fitting is calculated based on the optimized objective function value after optimization. Based on the dual-pulse parameters, delayed release features are extracted and an evidence vector is constructed, including: Hysteresis separation is obtained based on the relationship between the interval between the center velocities of the first pulse and the second pulse and the sum of the widths of the first pulse and the second pulse. The release steepening ratio is obtained based on the ratio of the amplitude to the width of the second pulse relative to the ratio of the amplitude to the width of the first pulse. The residual retention ratio of the later stage is obtained based on the residual response in the velocity domain after the center velocity of the second pulse plus twice the width of the second pulse. The hysteresis separation degree, the release steepening ratio, and the post-term residual retention ratio are combined to form an evidence vector; The formula for calculating the hysteresis separation degree is: ； in Indicates the degree of hysteresis separation. Indicates the center velocity of the first pulse. Indicates the center velocity of the second pulse. Indicates the width of the first pulse. Indicates the width of the second pulse. It represents a very small quantity to prevent the denominator from being zero; hysteresis separation is used to quantify the degree of separation of the two pulses on the velocity axis; The formula for calculating the steepening ratio is: ； in Indicates the release steepening ratio, This indicates the amplitude of the first pulse. The amplitude of the second pulse is indicated by the release steepness ratio, which characterizes the degree of energy increase of the second pulse compared to the first pulse. The formula for calculating the residual retention ratio in the later stage is: ； Where Q represents the residual retention ratio in the latter part. This represents the angular velocity value corresponding to the k-th sampling point. This represents the residual response value at the kth sampling point. The residual retention ratio in the latter part is used to quantify the degree of decline of the anomalous response after the occurrence of the double pulse. The formula for calculating the objective function is as follows: ； in This indicates the value of the objective function to be optimized. This represents the parameter vector to be optimized. This represents the angular velocity value corresponding to the k-th sampling point within the target's narrow velocity domain. Indicates the target narrow velocity range, This represents the residual response value at the k-th sampling point within the target narrow velocity domain. This represents the model fit value for the corresponding sampling point. This indicates a very small quantity that prevents the denominator from being zero; This is the fitting error term; These are structural constraint terms.

2. The intelligent fault early warning and decision-making method for a medical centrifuge according to claim 1, characterized in that, By combining the abnormal distance between the aforementioned evidence vector and the historical healthy operating baseline, a comprehensive early warning intensity is obtained, including: The historical healthy operating baseline includes the mean vector and covariance matrix under the same rotor type and similar loading conditions; Calculate the Mahalanobis anomaly distance between the evidence vector and the historical healthy operating baseline, and obtain the delayed release hazard probability based on the Mahalanobis anomaly distance; The overall warning intensity is obtained by multiplying the delayed release danger probability by the double pulse fitting confidence level.

3. The intelligent fault early warning and decision-making method for a medical centrifuge according to claim 2, characterized in that, Determining target control actions based on the comprehensive early warning intensity includes: The target control actions include continued acceleration, speed maintenance verification, and controlled shutdown; Based on the comprehensive early warning intensity and the residual retention ratio in the later stage, calculate the first utility value corresponding to continued acceleration, the second utility value corresponding to speed maintenance verification, and the third utility value corresponding to controlled shutdown. Compare the first utility value, the second utility value, and the third utility value to determine the target control action.

4. The intelligent fault early warning and decision-making method for a medical centrifuge according to claim 3, characterized in that, Controlling the medical centrifuge to perform the target control action includes: If the target control action is to continue accelerating, then maintain the current acceleration trajectory; If the target control action is speed-hold verification, the verification rotation speed is set to the center speed of the second pulse, and the verification duration is determined based on the sum of the first pulse width and the second pulse width and the current rate of change of angular velocity. The verification comprehensive warning intensity is recalculated during the verification phase. When the verification comprehensive warning intensity is lower than the original comprehensive warning intensity, the speed increase is resumed. When the verification comprehensive warning intensity is not lower than the original comprehensive warning intensity, a controlled shutdown is executed and a hoisting point check decision is output. If the target control action is controlled shutdown, then deceleration shutdown is executed, and a hysteresis release risk warning and basket hinge point inspection decision are output.

5. An intelligent fault early warning and decision-making system for a medical centrifuge, applied in the intelligent fault early warning and decision-making method for a medical centrifuge according to any one of claims 1-4, characterized in that, The system includes: The data construction module is used to collect vibration signals, current signals and speed signals during the acceleration phase of a medical centrifuge, perform equal-angle resampling based on the speed signals, and construct the acceleration composite response quantity. The interval identification module is used to perform background separation on the acceleration composite response, obtain the residual response, and determine the target narrow velocity range based on the residual response; The parameter inversion module is used to construct a double-pulse fitting model based on the residual response in the target narrow velocity domain, optimize the double-pulse fitting model using the particle swarm optimization algorithm, and obtain the double-pulse parameters and the double-pulse fitting reliability. The early warning assessment module is used to extract hysteresis release features based on the dual-pulse parameters and construct an evidence vector. Combined with the abnormal distance between the evidence vector and the historical healthy operation baseline, the comprehensive early warning intensity is obtained. The hysteresis release features include hysteresis separation degree, release steepening ratio and post-term residual retention ratio. The decision control module is used to determine the target control action based on the comprehensive early warning intensity and control the medical centrifuge to execute the target control action.

Images