CN-120492973-B - Infusion pump motor fault detection method based on self-adaptive weight SVM model

CN120492973BCN 120492973 BCN120492973 BCN 120492973BCN-120492973-B

Abstract

The invention discloses an infusion pump motor fault detection method based on a self-adaptive weight SVM model, which relates to the field of motor fault intelligent diagnosis, and comprises the following steps of obtaining a sample set; the method comprises the steps of calculating initial weights of different types of faults based on fault frequency and fault severity scores, constructing a motor fault detection SVM model, defining an objective function based on the initial weights, training the motor fault detection SVM model by utilizing a sample set, updating weights of corresponding types of faults based on misclassification loss, optimizing the objective function based on the updated weights, constructing a Lagrangian function, solving, inputting running characteristics of a to-be-detected infusion pump motor into the motor fault detection SVM model, and outputting detection results. According to the method, different initial weights are set for different types of faults, so that the accuracy of classification of the model on key faults can be improved. The invention adopts a self-adaptive weight mechanism, adjusts the weight according to the misclassification loss, and improves the generalization capability of the model.

Inventors

ZHAO YINGHONG
ZHONG MING
CHEN TIANRAN
ZHAO ZHIGANG
YANG PENG

Assignees

四川中测仪器科技有限公司

Dates

Publication Date: 20260508
Application Date: 20250507

Claims (8)

1. An infusion pump motor fault detection method based on an adaptive weight SVM model is characterized by comprising the following steps: Step one, acquiring historical data of operation characteristics of an infusion pump motor and corresponding operation states of the infusion pump motor to obtain a sample set; Calculating initial weights of different types of faults based on fault frequency and fault severity scores; Constructing a motor fault detection SVM model, and defining an objective function based on the initial weight; Training the motor fault detection SVM model by using the sample set, and calculating error classification loss of fault detection; Updating the weight of the corresponding type of faults based on the misclassification loss; Step six, optimizing the objective function based on the updated weight, constructing a Lagrangian function and solving to obtain a classified hyperplane parameter; Inputting the operation characteristics of the infusion pump motor to be detected into the motor fault detection SVM model, and outputting a fault detection result of the infusion pump motor to be detected according to the classification hyperplane parameters; In the first step, the operation characteristics of the motor of the infusion pump comprise motor output pressure and motor phase current effective ripple characteristics, and the motor output pressure is obtained through the following steps: The pressure sensor is used for collecting the actual measured pressure of the pipeline liquid in real time; obtaining a vertical height difference between an infusion pump and an infusion part of a patient, and calculating based on the vertical height difference and the liquid medicine density to obtain compensation pressure; calculating to obtain the motor output pressure based on the measured pressure and the compensation pressure; The motor phase current effective ripple characteristic is obtained through the following steps: monitoring the dynamic change of the motor output pressure, and when the motor output pressure is suddenly changed, calculating the change rate of the motor output pressure along with time in a sudden change period to generate a pressure change rate sequence; Synchronously collecting current waveform data of the motor winding of the infusion pump in the abrupt change period, calculating and obtaining current fluctuation rate in each driving period based on the current waveform data, and generating a current fluctuation rate sequence; Calculating a time domain correlation coefficient of the pressure change rate sequence and the current fluctuation rate sequence, and marking the current fluctuation rate sequence as the motor phase current effective ripple characteristic if the time domain correlation coefficient exceeds a preset coefficient threshold value.
2. The method for detecting the failure of the infusion pump motor based on the adaptive weight SVM model according to claim 1, wherein in the first step, the operation characteristics of the infusion pump motor further comprise dynamic characteristics of the infusion part of the patient, and the dynamic characteristics of the infusion part of the patient are obtained by the following steps: acquiring the displacement duration time of the infusion part of the patient and the three-dimensional space displacement acceleration of the infusion part of the patient in real time; Calculating and obtaining the instantaneous displacement of the transfusion part of the patient based on the three-dimensional space displacement acceleration; and if the displacement duration is lower than a preset time threshold and the instantaneous displacement exceeds the preset displacement threshold, marking the displacement duration and the instantaneous displacement as dynamic characteristics of the transfusion part of the patient.
3. The method for detecting faults of an infusion pump motor based on an adaptive weight SVM model according to claim 1, wherein in the second step, the calculating initial weights of different types of faults based on fault frequency and fault severity score comprises: calculating a failure frequency based on the history; Evaluating the severity of the fault based on the threat level of the fault to the safety of the patient, the maintenance cost level and the shutdown time period level of the infusion pump motor, and obtaining the fault severity score; The initial weight of the fault is calculated according to the following formula: ; Wherein, the As an initial weight for a class i fault, As the failure frequency of the i-th type of failure, The fault severity for the i-th class of fault is scored.
4. The method for detecting a failure of an infusion pump motor based on an adaptive weight SVM model according to claim 3, wherein in the third step, the objective function is: ; Wherein, the As the weight coefficient of the light-emitting diode, In order for the offset to be a function of, In order to relax the variables of the variables, In order for the parameters to be regularized, For the number of samples to be taken, As a feature vector for a class i fault, As an output value for a class i fault, To classify hyperplane equations.
5. The method for detecting faults of an infusion pump motor based on an adaptive weight SVM model according to claim 4, wherein in the fourth step, a calculation formula of misclassification loss of fault detection is as follows: ; Wherein, the Misclassification loss for class i faults.
6. The method for detecting faults of an infusion pump motor based on an adaptive weight SVM model according to claim 5, wherein in the fifth step, the formula for updating the weight of the corresponding type of faults based on the misclassification loss is as follows: ; Wherein, the In order for the rate of learning to be high, For the number of iterations, Is the weight of the i-th class of fault.
7. The method for detecting faults of an infusion pump motor based on an adaptive weight SVM model according to claim 6, wherein in the sixth step, a Lagrangian function is constructed as follows: ; Wherein, the Is a lagrange multiplier.
8. The method for detecting faults of an infusion pump motor based on an adaptive weight SVM model according to claim 7, wherein in the seventh step, the fault decision classification expression for outputting the fault detection result of the infusion pump motor to be detected according to the classification hyperplane parameters is: ; Wherein, the A kernel of an SVM model is detected for the motor fault, , As a parameter of the kernel function, Is the characteristic vector of the infusion pump motor to be detected.

Description

Infusion pump motor fault detection method based on self-adaptive weight SVM model Technical Field The invention relates to the field of intelligent diagnosis of motor faults, in particular to an infusion pump motor fault detection method based on a self-adaptive weight SVM model. Background Infusion pump is an instrument which can accurately control the number of drops of infusion or the flow rate of infusion, ensure the uniform speed of the drug, ensure the accurate dosage and safely enter the body of a patient to play a role, and is commonly used for the situation that the infusion amount and the dosage need to be strictly controlled, such as the application of boosting drugs, antiarrhythmic drugs and infant intravenous infusion or intravenous anesthesia. One of the core components of the infusion pump is a stepping motor for driving the pump system to work, and the stepping motor can accurately control the infusion speed and the infusion quantity by controlling the rotation speed and the rotation times of the stepping motor. When the infusion pump motor fails, the infusion pump motor must be discovered and processed in the first time, however, the existing failure detection method mainly depends on threshold judgment or simple statistical analysis, and has limitations in detecting complex failures. The prior art CN111474476a discloses a motor fault prediction method, which mainly relies on an SVM model to perform fault diagnosis on a motor, but the method does not consider the problem of unbalance of a sample data set of motor faults, i.e. the number of samples of certain types of faults is far greater than that of other types of faults, and the same punishment degree is set for all fault types. This results in that the classifier may tend to classify the samples to be tested into fault types with a large number of samples, since the model is more affected by the majority class samples during the training process, while ignoring the features of the minority class samples. Especially in the face of faults that occur less frequently (with a small number of samples) but are very dangerous, the model is difficult to identify accurately. Disclosure of Invention The invention aims to improve the accuracy of the motor fault detection SVM model for the infusion pump motor fault detection. In order to achieve the above purpose, the invention provides a method for detecting faults of an infusion pump motor based on an adaptive weight SVM model, which comprises the following steps: Step one, acquiring historical data of operation characteristics of an infusion pump motor and corresponding operation states of the infusion pump motor to obtain a sample set; Calculating initial weights of different types of faults based on fault frequency and fault severity scores; Constructing a motor fault detection SVM model, and defining an objective function based on the initial weight; Training the motor fault detection SVM model by using the sample set, and calculating error classification loss of fault detection; Updating the weight of the corresponding type of faults based on the misclassification loss; Step six, optimizing the objective function based on the updated weight, constructing a Lagrangian function and solving to obtain a classified hyperplane parameter; And step seven, inputting the operation characteristics of the infusion pump motor to be detected into the motor fault detection SVM model, and outputting a fault detection result of the infusion pump motor to be detected according to the classification hyperplane parameters. The Support Vector Machine (SVM) algorithm is a data mining method based on statistical learning theory, and the mechanism is to find a classification hyperplane meeting classification requirements, so that optimal classification of linear separable data can be realized theoretically. There are many possible types of faults for the infusion pump motor, such as rotation faults, noise faults, temperature faults, etc., and the occurrence frequency and severity of these different types of faults are different. In order to enable the model to learn the boundary of the key faults, when the objective function of the motor fault detection SVM model is defined, different initial weights are set for different types of faults according to the fault frequency and the fault severity score, so that different types of faults have different punishment degrees, the model is forced to be capable of considering faults with low sample number effectively, the weights are reversely adjusted according to the misclassification loss in the subsequent training process of the model, the weights are increased for continuous misclassification faults, and the classification boundary is optimized. And finally, optimizing an objective function according to the updated weight, constructing a Lagrange function solution, and obtaining a decision boundary capable of maximizing the classification interval and minimizing the weig