CN114894289B - Large-mass comparator based on data fusion algorithm - Google Patents
Large-mass comparator based on data fusion algorithm Download PDFInfo
- Publication number
- CN114894289B CN114894289B CN202210694262.7A CN202210694262A CN114894289B CN 114894289 B CN114894289 B CN 114894289B CN 202210694262 A CN202210694262 A CN 202210694262A CN 114894289 B CN114894289 B CN 114894289B
- Authority
- CN
- China
- Prior art keywords
- data
- sensor
- weighing
- value
- sensors
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000004927 fusion Effects 0.000 title claims abstract description 35
- 238000005303 weighing Methods 0.000 claims abstract description 56
- 238000000034 method Methods 0.000 claims abstract description 32
- 238000013528 artificial neural network Methods 0.000 claims abstract description 24
- 238000001914 filtration Methods 0.000 claims abstract description 23
- 230000003044 adaptive effect Effects 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 35
- 238000006243 chemical reaction Methods 0.000 claims description 20
- 238000004891 communication Methods 0.000 claims description 14
- 238000012549 training Methods 0.000 claims description 13
- 238000012937 correction Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000003491 array Methods 0.000 claims description 2
- 210000002569 neuron Anatomy 0.000 claims 2
- 238000005516 engineering process Methods 0.000 claims 1
- BTCSSZJGUNDROE-UHFFFAOYSA-N gamma-aminobutyric acid Chemical compound NCCCC(O)=O BTCSSZJGUNDROE-UHFFFAOYSA-N 0.000 claims 1
- 238000010606 normalization Methods 0.000 claims 1
- 238000005070 sampling Methods 0.000 description 5
- 230000003247 decreasing effect Effects 0.000 description 3
- 230000035945 sensitivity Effects 0.000 description 3
- 208000000044 Amnesia Diseases 0.000 description 2
- 208000031091 Amnestic disease Diseases 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 2
- 230000006986 amnesia Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 230000002596 correlated effect Effects 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000001105 regulatory effect Effects 0.000 description 2
- 239000010959 steel Substances 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01G—WEIGHING
- G01G23/00—Auxiliary devices for weighing apparatus
- G01G23/01—Testing or calibrating of weighing apparatus
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01G—WEIGHING
- G01G3/00—Weighing apparatus characterised by the use of elastically-deformable members, e.g. spring balances
- G01G3/12—Weighing apparatus characterised by the use of elastically-deformable members, e.g. spring balances wherein the weighing element is in the form of a solid body stressed by pressure or tension during weighing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03H—IMPEDANCE NETWORKS, e.g. RESONANT CIRCUITS; RESONATORS
- H03H21/00—Adaptive networks
- H03H21/0012—Digital adaptive filters
- H03H21/0025—Particular filtering methods
- H03H21/0029—Particular filtering methods based on statistics
- H03H21/003—KALMAN filters
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Testing Or Calibration Of Command Recording Devices (AREA)
Abstract
The invention discloses a large mass comparator based on a data fusion algorithm, which comprises a mass comparator scale body and an instrument; the instrument is provided with an initialization module for initializing, and comparing the obtained weighing sensor data with pre-stored horizontal state data according to real time to determine whether the weighing sensor data are in a horizontal state or not; the fault identification module is used for fusing the data of the symmetrical weighing sensors and identifying the weighing sensor with the fault according to the correlation among the sensors; an adaptive filtering module for adaptively determining a Kalman filter by standard deviations of a plurality of sensor dataRValue sumQA value; and a data fusion module: the method comprises the steps of taking multipath weighing sensor signals as input, establishing a quality comparator error compensation model by using an error compensation method of radial basis function neural network multi-sensor information fusion, and taking the compensated value as outputThe invention can solve the problems of scale body horizontal adjustment, unbalanced load error and nonlinear error of the large-mass comparator in actual use.
Description
Technical Field
The invention belongs to the field of mass measurement, and particularly relates to a large-mass comparator based on a data fusion algorithm.
Background
The mass comparator is widely applied to mass measurement, is a special electronic balance, is mainly used for mass comparison of standard weights and checked weights, and is a very important magnitude transmission tool. The main tracing principle is that the measured weight and the standard weight are compared for a plurality of times through the mass comparator to obtain the difference value of the measured weight and the standard weight, and the value of the measured weight can be obtained due to the mass of the known standard weight. Currently, mass comparators are mainly classified into two types, electromagnetic force sensors and strain sensors. The electromagnetic force sensor has high precision, is mainly used for the magnitude transmission of weights with small mass and high precision (E and F grades), has high requirements on the use environment, and can not move once the installation and the debugging are finished. The mass comparator of the strain sensor is mainly used for transmitting the magnitude of the weight for large-mass (more than 1 t) and low-precision (M grade) work, the requirement on the use environment is relatively low, the strain sensor can be used by adjusting the level after moving, and the mass comparator is simply called a large-mass comparator and plays a very important role in the aspects of weighing apparatus manufacture, traffic safety and the like.
The currently used large-mass comparator mostly adopts a type of 3 or 4 sensors, and all sensor signals are connected in parallel and then are converted into mass values after being processed by signal amplification, filtering and the like. The problems are mainly:
1. it is difficult to adjust the level of the balance body. The large mass comparator must be adjusted horizontally after moving, so that the sensor can be normally used only when being positioned at the same horizontal plane, and at present, most screws on each angle are adjusted to be positioned at the horizontal position by adopting a mode of observing horizontal bubbles by naked eyes, so that the mode has larger visual observation error and is difficult to adjust to the horizontal position.
2. Unbalanced load errors and nonlinear errors of the large-mass comparator are difficult to solve. All sensors of the mass comparator are connected in parallel to form a signal through a junction box and then are subjected to AD conversion, and because parameters such as sensitivity and the like of each sensor are different, the potentiometer on the junction box needs to be adjusted before the signals of each sensor are connected in parallel, so that the sensitivity is consistent as much as possible, the weight is placed at different positions to display consistent mass values, the problem of unbalanced load error is solved, the actual adjustment operation is very difficult, and the manual adjustment of the potentiometer is time-consuming and labor-consuming. Meanwhile, nonlinear errors generated by the problems of deformation, stress and the like of the scale body cannot be solved at present, and the precision of the large-mass comparator is affected.
Because of the defects, the use effect of many large-mass comparators in the practical application process is not ideal, and the manual adjustment is difficult, so that the requirements of on-site use of clients are difficult to meet.
Disclosure of Invention
The invention aims to solve the problems of scale body horizontal adjustment, unbalanced load error and nonlinear error in actual use of a large-mass comparator based on a data fusion algorithm.
The technical solution for realizing the purpose of the invention is as follows:
a large mass comparator based on a data fusion algorithm comprises a mass comparator body provided with a plurality of paths of weighing sensors; the mass comparator scale body is connected with the instrument through a junction box; the meter is provided with:
the initialization module is used for completing initialization of data parameters, and comparing the obtained weighing sensor data with pre-stored level state data according to real time so as to determine whether the quality comparator scale body is in a level state or not;
the fault identification module is used for fusing the data of the weighing sensors based on the Pearson correlation coefficient and identifying the weighing sensors with faults according to the correlation among the sensors;
an adaptive filtering module, based on standard deviation adaptive Kalman filtering, adaptively determining Kalman filtering by standard deviation of a plurality of sensor dataRValue sumQA value;
a data fusion module based on a radial basis function neural network: the method comprises the steps of taking multiple paths of weighing sensor signals as input, establishing a mass comparator error compensation model by using an error compensation method of radial basis function neural network multi-sensor information fusion, and obtaining a weighing result by taking the compensated value as output.
Compared with the prior art, the invention has the remarkable advantages that:
(1) According to the invention, the AD conversion is carried out on each strain weighing sensor in a digital junction box mode, and the coefficient of each sensor is automatically calculated in an artificial intelligence mode, so that the sensitivity of the sensor is prevented from being adjusted by adopting a potentiometer, and the method is more accurate and convenient.
(2) The invention designs a level adjustment function, whether the weighing sensor is in a level state is judged by comparing the value of each weighing sensor with the value of the weighing sensor in the level state when the weighing sensor is in no-load state, and the height of each angle can be adjusted according to the value of each weighing sensor so as to achieve the level state.
(3) The invention designs a data fusion algorithm, which is used for carrying out fault identification on a sensor through the data fusion algorithm based on the Pearson correlation coefficient, judging whether the sensor is abnormal or not, and carrying out data fusion on four sensors through a radial basis function neural network after self-adaptive Takalman filtering so as to effectively solve unbalanced load errors and nonlinear errors.
Drawings
Fig. 1 is a block diagram of a mass comparator based on a data fusion algorithm.
FIG. 2 is a flow chart of an initialization module.
FIG. 3 is a flow chart of a data fusion algorithm.
Fig. 4 is a radial basis function neural network graph.
Detailed Description
For the purpose of illustrating the technical scheme and technical purposes of the present invention, the present invention is further described below with reference to the accompanying drawings and specific embodiments.
Referring to fig. 1 to 4, a mass comparator based on data fusion according to the present invention is mainly divided into three parts: the mass comparator comprises a mass comparator body, an embedded digital junction box and an instrument. The mass comparator scale body consists of four sensors and a steel structure table top. The embedded digital junction box mainly comprises an AD conversion module and a communication module. The instrument mainly comprises an initialization module, a data fusion module and a display module.
The mass comparator scale body is connected with the AD conversion module; the AD conversion module is connected with the communication module; the communication module is connected with an initialization module of the instrument, the initialization module is a module executed when the instrument is just started up for use, and once the operation is finished, the module is not required to be executed in the use process. In the use process, the communication module is directly connected with the data fusion module, and the data fusion module is connected with the display module.
The weight comparator scale body is used for bearing weights and converting the mass of the weights into voltage signals.
The working process of the mass comparator scale body is as follows:
the weighing platform surface is a steel platform with the thickness of 1.2m multiplied by 1.2m, and four weighing sensors are respectively arranged on four feet. When the weight to be detected is placed on a table top, the four weighing sensors are stressed, so that the mass of the weight is converted into four paths of voltage signals, and the four paths of voltage signals are output to a digital-to-analog conversion module (AD conversion module for short) of the embedded junction box in parallel.
The AD conversion module is used for converting the voltage signal of the weighing sensor into a digital signal so as to analyze and process the later-stage signal.
The AD conversion module comprises the following working processes:
the AD conversion module consists of four AD7176 data conversion plates, and each AD data conversion plate is connected with a weighing sensor. Each AD data conversion plate sequentially circularly collects signals of each weighing sensor according to preset settings, and data of four sensors are transmitted to the communication module after each cycle is collected so as to be transmitted to the upper computer. The STM32F103 singlechip in the embedded junction box is responsible for setting the sampling mode, the sampling rate and the like of the AD7176 conversion plate. In order to improve the accuracy of AD conversion, the sampling rate of the four AD7176 data conversion plates is set to 16 times/second, the sampling mode is set to be circulation sampling, and when the quality comparator is powered on and begins to be used each time, the STM32F103 singlechip in the embedded junction box firstly carries out parameter setting on the AD7176 data conversion module, and parameters can not be changed any more in the whole use process.
The communication module is used for transmitting the data converted by the AD7176 conversion board to an upper computer instrument in real time.
The working process of the communication module is as follows:
the communication module adopts an RS422 communication mode, firstly judges whether the data of the four paths of sensors are complete, if the data of the four paths of sensors are complete, the data of the four slave sensors are respectively sent in sequence, and a data head and a data tail are added so that the upper computer instrument can analyze the data. The specific data format is as follows:
AA ×× ×× ×× ×× ×× ×× ×× ×× ×× ×× ×× ×× FF
AA is a data header and a set of data starts to be identified. FF is the tail of the data, a set of end of data flags. The middle is the data of the weighing sensors, and each sensor occupies 24 bits.
The communication module sends the sensor data to the instrument, if the instrument is just started, the initialization flow is needed, the communication data is transmitted to the initialization module, and after the initialization is finished, the initialization module is not needed to be executed.
The working process of the initialization module is as follows:
as shown in fig. 2, the initialization module mainly initializes the data parameters of the instrument and adjusts the level of the quality comparator.
The initialization parameters are mainly parameters for training a radial basis function neural network, and the balance body of the quality comparator pre-stores level state sensor data. The parameters of the radial basis function neural network training are mainly used for data fusion of four sensors. The sensor data of the level state of the balance body of the quality comparator refers to the values of four sensors pre-stored in the level state of the balance body [ ]sensor1, sensor2, sensor3, sensor4) This data is obtained by pre-calibration.
sensor1 refers to the number value of the sensor No. 1 in the horizontal state;
sensor2 refers to the horizontal stateNumber 2 sensor;
sensor3 refers to the number value of the sensor No. 3 in the horizontal state;
sensorthe number 4 refers to the number 4 sensor in the horizontal state;
when the weighing scale is initialized, the weighing scale body is kept in an idle state, and after the initialization of the initialization module is completed, the data of the four weighing sensors are read through the serial portreal1, real2, real3, real4)。
real1 refers to data read by a sensor 1 in real time;
real2 refers to data read by a sensor No. 2 in real time;
real3 refers to data read by a sensor No. 3 in real time;
real4 refers to data read by a sensor No. 4 in real time;
then obtaining the data of the weighing sensor in real timereal1, real2, real3, real4) And the pre-stored level state datasensor1, sensor2, sensor3, sensor4) Comparing, if the sensor data read in real time is smaller, the sensor data indicates that the angular position of the sensor is lower; if the sensor data read in real time is larger, the position of the angle where the sensor is located is higher. Simultaneously satisfies:
the balance body of the quality comparator is in a horizontal state, so that the normal use requirement is met. If the indication that the quality comparator body is not in a horizontal state cannot be met, the instrument displays alarm information. A is a horizontal state judgment threshold, 1000 is taken in the embodiment, and the adjustment can be performed according to different sensors.
After initialization is completed, the large-mass comparator enters a normal working state, and the data fusion module reads the data of the four weighing sensors. The data fusion module workflow is shown in fig. 3.
The data fusion module is mainly divided into three parts: the system comprises a fault identification module based on pearson correlation coefficient data fusion, a Kalman self-adaptive filtering module and a data fusion module based on a radial basis function neural network.
Firstly, data of four sensors read in by a serial port are subjected to fault identification based on data fusion of Pearson correlation coefficients, and the specific idea is as follows:
in the process of loading and unloading weights, the change rules of the four sensors are the same, the data should be increased or decreased at the same time, the relation between every two of the four sensors is strongly correlated, and if the data of the weighing sensor is not strongly correlated with the data of other weighing sensors, the sensor is indicated to have a fault and is not in a normal working state, and adjustment or maintenance is needed.
The method comprises the following steps of
1) The four sensor data are respectively stored in an array of 20 data respectively、/>、、/>The standard deviation of each array was calculated separately using equation (1), and recorded as +.>：
（1）
Wherein the method comprises the steps ofFor corresponding sensor arrayFirst, thelThe data of the plurality of data,l=1,2，…20；/>is the firstiStandard deviation of individual sensors, when +.>Get->When (I)>Standard deviation for sensor 1; similarly, when->Get->When (I)>Standard deviation for sensor 2;
an average value of 20 data in the corresponding sensor array;
nto correspond to the number of data in the sensor arrayn=20
2) Then, the Pearson correlation coefficient between every two sensors is calculated by using the formula (2)
（2）
In the middle ofXAndYtwo arrays of sensors for which pearson correlation coefficients need to be calculated.
Is thatXStandard deviation of>Is thatYStandard deviation of (2);
cov in the middleX，Y) Is thatXAndYis a covariance of (2);
Eis a mathematical expectation.
3) If it isAnd the weight sensor is smaller than a first set threshold B1, in this embodiment, the weight sensor B1 takes 100, which indicates that the weight body is in an empty load state or the weight is placed completely, the weight sensor should be in an irregular fluctuation state, and the four sensors are determined to be irrelevant, and the pearson correlation coefficient is not calculated at this time, so that fault identification is not performed.
4) If it isIf the sensor data is greater than or equal to the first set threshold B1, indicating that the sensor data is in a severe change, namely in a loading and unloading state, determining that the four sensors are strongly related at the moment, and judging that the four sensors are strongly related at the moment>The value of (2) should be between 0.5 and 1. If the value is less than 0.5 or negative, the correlation of the sensor is not strong, so that the weighing sensor with faults can be analyzed and identified.
5) Reading the next data repeats the above process.
Adaptive Kalman filtering based on standard deviation if no fault occurs, by standard deviation of four sensor dataAdaptively determining Kalman filteringRValue sumQValues.
Covariance parameters of Kalman filteringRValue sumQThe values affect the convergence speed of the filtering and the filtering model error. In order to ensure the convergence rate and obtain a numerical value with small model error, the large-mass comparator adopts a Kalman self-adaptive filtering algorithm based on standard deviation. The main idea is as follows:
when the weight is loaded or unloaded, the data of the weighing sensor can be rapidly increased or decreased, and the filtering following performance is required to be better,Rvalue sumQThe value may take a larger value in order to save convergence time; after no-load or weight placement is completed, the data after filtering can be kept stable, relevant interference factors are filtered, so as to obtain stable and accurate values, and a smaller value is required to be setRValue sumQValues. The standard deviation of the data can precisely reflect the change rate of the data, and is very large when the data of the loading and unloading weights are increased or decreased sharply; after the weight or no load stabilizes, the standard deviation of the load cell data will be smaller, and the adjustment is made accordinglyRValue sumQValues.
The method comprises the following specific steps:
1) Reading in the standard deviation of the four sensor data calculated above；
2) If it isAnd the weight is larger than or equal to a first set threshold B1 (100), which indicates that the weight is not subjected to Kalman filtering in the loading or unloading process.
If it isRespectively smaller than a first set threshold B1 (100) and larger than or equal to a second set threshold B2 (80), which indicates that the loading or unloading of the weight is completed, but the scale body is in intense stateVibration state, then set up:
if it isAnd the weights are respectively smaller than a second set threshold B2 (80) and larger than or equal to a third set threshold B3 (50), which indicates that the loading or unloading of the weights is completed, but the balance body is in a larger vibration state, and then:
if it isAnd the weight is respectively smaller than a third set threshold B3 (50) and larger than or equal to a fourth set threshold B4 (30), after the weight loading or unloading is finished, the balance body is in a smaller vibration state, and then:
if it isRespectively smaller than a fourth set threshold B4 (30), after the weight is loaded or unloaded, the balance body is basically stable, and then:
3) And (3) each sensor array queue re-reads new data as a queue head, discards the last data at the tail of the queue, and repeats 1) for circular filtering.
The filtered data is mainly used for solving the problem of unbalanced load error of the quality comparator based on a radial basis function neural network algorithm. The main idea is as follows: the output of the four-way sensor of the traditional quality comparator adopts a parallel circuit mode, the potentiometer in the junction box is repeatedly regulated, and the gain of each channel is regulated, so that the unbalanced load error is reduced. However, the unbalanced load error of the mass comparator is influenced by nonlinear factors such as the rigidity and the strength of the scale body, the internal stress in the processing and mounting process, the mechanical deformation of the scale body during bearing and the like, and the required precision can be achieved by only adjusting the potentiometer, so that the unbalanced load error is an important factor influencing the precision of the mass comparator. The radial basis function neural network has wide application in multi-sensor information fusion, and can well approximate to a nonlinear function. The algorithm takes four paths of weighing sensor signals as input, establishes a mass comparator error compensation model by using an error compensation method of multi-sensor information fusion of a radial basis function neural network, takes the compensated value as output, and the compensation model is shown in the figure 4, wherein:
（3）
wherein:to the compensated output value. Number of hidden layer neuronsmExperiments prove that the comparator is subjected to a plurality of experiments,m=20；Wweight vector for radial basis function neural network, +.>Wherein->；HFor radial basis function vector +.>Wherein->；dIs the input layer bias value.
The algorithm adopts a Gaussian function as a basis function of a network, and:
（4）
wherein:is an input vector, wherein->，/>Respectively the firsti（iValues of =1, 2,3, 4) sensors; />Is the sensor ofjExtension constants of the individual nodes; />Is the sensor ofjThe center vector of the individual nodes is used,，/>respectively the firsti（i=1, 2,3, 4) sensor numberjCenter vector of each node.
The method comprises the following specific steps:
firstly, training an error compensation basis function neural network of a quality comparator. The quality comparator uses 500kg,1t and 2t three numerical values to load different positions of the balance body of the quality comparator respectively by utilizing different masses, and 60 groups of 4 paths of weighing sensor signals are acquired and normalized to obtain dataIs used for training the basis function neural network. Training the radial basis function by adopting a gradient training method, wherein the learning objective function is as follows:
（5）
in the middle ofIs amnesia factor, is->Is an error signal.
（6）
Wherein the method comprises the steps ofThe true value is the mass of the weight.
To minimize using the objective function, the correction of each parameter should be proportional to its negative gradient. The respective parameter correction amounts can be obtained:
（7）
（8）
（9）
（10）
wherein the method comprises the steps ofIs the learning rate.
Objective functionESet to 1X 10 -8 Learning rateAre all 0.015, amnesia factor +.>Both 0.6. Training according to formula (5), formula (6), formula (7), formula (8), formula (9) and formula (10) by using 60 groups of data to obtain parameter ∈ ->Center vector->Weight matrixWAnd output layer bias +.>Is a value of (2). The process only needs to collect and train data when the quality comparator is used for the first time, once the basis function neural network is trained, the training is not needed, and parameters are directly used in the later period.
The display module is mainly used for displaying relevant fault information and weighing results in real time.
According to the invention, the four sensor data of the large-mass comparator are analyzed and processed through the initialization module, so that the horizontal adjustment of the large-mass comparator can be conveniently completed, the influence of unbalanced load and nonlinear errors can be reduced through a data fusion algorithm, and the precision of the large-mass comparator is improved.
Claims (3)
1. A large mass comparator based on a data fusion algorithm comprises a mass comparator body provided with a plurality of paths of weighing sensors; the mass comparator is characterized in that the balance body is connected with an instrument through a junction box; the meter is provided with:
the initialization module is used for completing initialization of data parameters, and comparing the obtained weighing sensor data with pre-stored level state data according to real time so as to determine whether the quality comparator scale body is in a level state or not;
the fault identification module is used for fusing the data of the weighing sensors based on the Pearson correlation coefficient and identifying the weighing sensors with faults according to the correlation among the sensors;
the self-adaptive filtering module is used for self-adaptively determining R value and Q value of the Kalman filtering through standard deviation of a plurality of sensor data based on self-adaptive Kalman filtering of the standard deviation;
a data fusion module based on a radial basis function neural network: taking multiple paths of weighing sensor signals as input, and establishing a quality comparator error compensation model by using an error compensation method of radial basis function neural network multi-sensor information fusion, wherein the compensated value is taken as output to obtain a weighing result;
the engineering process of the initialization module is as follows:
when the weighing scale is initialized, the weighing scale body is kept in an idle state, and after the initialization of data parameters is completed, the data of a plurality of weighing sensors are read;
then, weighing sensor data are obtained in real time and compared with pre-stored horizontal state data, and if the sensor data read in real time are smaller, the sensor is indicated to be positioned at a lower angular position; if the sensor data read in real time are bigger, the position of the angle where the sensor is located is higher;
simultaneously satisfies:
|reali-sensori|≤A
the balance body of the quality comparator is in a horizontal state, so that the normal use requirement is met, and otherwise, the use requirement is not met;
wherein reali is the data read by the ith weighing sensor in real time, and sensor is the value of the sensor prestored by the ith weighing sensor in a horizontal state; a is a horizontal state judgment threshold;
the specific process of the fault identification module for identifying the weighing sensor with the fault comprises the following steps:
the sensor data are respectively stored in an array, and the standard deviation of each array is respectively calculated:
wherein x is l Is the first data in the corresponding sensor array; s is S i Standard deviation for the ith sensor;an average value of the corresponding sensor array data; n is the number of data in the corresponding sensor array;
the pearson correlation coefficient between the two sensors is then calculated:
wherein X and Y are two sensor arrays for which the Pearson correlation coefficient needs to be calculated; sigma (sigma) x Is the standard deviation of X, sigma y Is the standard deviation of Y; cov (X, Y) is the covariance of X and Y; e is a mathematical expectation;
if S i When the weight is smaller than the set threshold, the balance body is in an empty load state or the weight is placed completely, and the four sensors are judged to be irrelevant, at the moment, the Pearson correlation coefficient is not calculated, and fault identification is not carried out;
if S i If the correlation coefficient between every two sensors is smaller than the set value, the correlation of the sensors is not strong, and accordingly the weighing sensor with faults is identified;
the adaptive filtering module determines that the R value and the Q value of the Kalman filter comprise the following working conditions:
if the standard deviation of each sensor is larger than or equal to a first set threshold value, indicating that the weight is not subjected to Kalman filtering in the loading or unloading process; and different R values and Q values are set according to the following different working conditions:
(1) If the standard deviation of each sensor is respectively smaller than the second set threshold value and larger than or equal to the third set threshold value B2, the weight loading or unloading is finished, but the balance body is in a larger vibration state;
(2) If the standard deviation of each sensor is respectively smaller than the third set threshold value and larger than or equal to the fourth set threshold value B2, the balance body is in a smaller vibration state after the weight loading or unloading is finished;
(3) If the standard deviation of each sensor is respectively smaller than a fourth set threshold value, indicating that the loading or unloading of the weight is completed;
the data fusion module based on the radial basis function neural network comprises the following steps:
firstly, training an error compensation basis function neural network of a quality comparator, respectively loading different positions of a balance body of the quality comparator by utilizing different qualities, collecting a plurality of groups of multipath weighing sensor signals, and obtaining data X after normalization processing k Training a basis function neural network;
training the radial basis function by adopting a gradient training method, wherein the learning objective function is as follows:
in eta i E is forgetting factor i Is an error signal;
correction amounts of the respective parameters:
setting an objective function E and a learning rate u i Forgetting factor eta i Training by using multiple groups of data to obtain expansion constant r of jth node of sensor j Center vector C of jth node of sensor j Weight matrix W and output layer bias d j Is a value of (2); w (w) j The j-th vector of the weight matrix W;
each time after completing the acquisition and filtering processing of the signals of the multipath weighing sensor, the signal data and each parameter r of the radial basis function neural network are processed j 、C j 、W、d j Substituting the compensation model and the basis function of the neural network respectively, and calculating to obtain a compensated output value y k Outputting the weighing result of the mass comparator in real time;
the basis functions of the compensation model and the neural network are respectively as follows:
and (3) compensating a model:
wherein: m is the number of hidden layer neurons, and W is the weight vector of the base neural network, wherein W is the number of hidden layer neurons 0 =1; h is a radial basis function vector; d is the input layer bias value;
basis functions:
j＝1，2，…，m
wherein: x is X k For an input vector, including the value of the sensor; r is (r) j Is an expansion constant of the jth node of the sensor.
2. The mass comparator based on the data fusion algorithm according to claim 1, wherein the junction box is composed of an AD conversion module and a communication module;
the AD conversion module is used for converting the voltage signal of the weighing sensor into a digital signal;
the communication module is used for sending the converted data to the upper computer instrument in real time.
3. The mass comparator based on the data fusion algorithm of claim 2, wherein the working process of the communication module is as follows:
firstly judging whether the data of the multi-path sensor is complete, if so, respectively transmitting the data of a plurality of slave sensors in sequence, and adding a data head and a data tail, wherein the specific data format is as follows:
AA ×× ×× ×× ×× ×× ×× ×× ×× ×× ×× ×× ×× FF
AA is a data head, and a group of data starts to be identified; FF is the data tail, a group of data end identifiers; the middle is the data of the weighing sensors, and each sensor occupies 24 bits.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210694262.7A CN114894289B (en) | 2022-06-20 | 2022-06-20 | Large-mass comparator based on data fusion algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210694262.7A CN114894289B (en) | 2022-06-20 | 2022-06-20 | Large-mass comparator based on data fusion algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114894289A CN114894289A (en) | 2022-08-12 |
| CN114894289B true CN114894289B (en) | 2024-02-02 |
Family
ID=82727607
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210694262.7A Active CN114894289B (en) | 2022-06-20 | 2022-06-20 | Large-mass comparator based on data fusion algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114894289B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117782289B (en) * | 2024-02-26 | 2024-05-14 | 深圳市杰曼科技股份有限公司 | Digital weighing system based on Internet of things and weighing method thereof |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104330084A (en) * | 2014-11-13 | 2015-02-04 | 东南大学 | Neural network assisted integrated navigation method for underwater vehicle |
| KR101907589B1 (en) * | 2018-01-22 | 2018-10-12 | 연세대학교 산학협력단 | Structural system identification using extended kalman filter and genetic algorithm |
| WO2018198576A1 (en) * | 2017-04-28 | 2018-11-01 | 株式会社Ｓｏｋｅｎ | Torque detection device and sensor module |
| CN110954132A (en) * | 2019-10-31 | 2020-04-03 | 太原理工大学 | Method for carrying out navigation fault identification through GRNN (generalized regression neural network) assisted adaptive Kalman filtering |
| CN111259927A (en) * | 2020-01-08 | 2020-06-09 | 西北工业大学 | Rocket engine fault diagnosis method based on neural network and evidence theory |
| WO2021018160A1 (en) * | 2019-07-31 | 2021-02-04 | Mettler-Toledo (Changzhou) Measurement Technology Ltd. | Hysteresis compensation method for a weighing device |
| CN112651456A (en) * | 2020-12-31 | 2021-04-13 | 遵义师范学院 | Unmanned vehicle control method based on RBF neural network |
-
2022
- 2022-06-20 CN CN202210694262.7A patent/CN114894289B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104330084A (en) * | 2014-11-13 | 2015-02-04 | 东南大学 | Neural network assisted integrated navigation method for underwater vehicle |
| WO2018198576A1 (en) * | 2017-04-28 | 2018-11-01 | 株式会社Ｓｏｋｅｎ | Torque detection device and sensor module |
| KR101907589B1 (en) * | 2018-01-22 | 2018-10-12 | 연세대학교 산학협력단 | Structural system identification using extended kalman filter and genetic algorithm |
| WO2021018160A1 (en) * | 2019-07-31 | 2021-02-04 | Mettler-Toledo (Changzhou) Measurement Technology Ltd. | Hysteresis compensation method for a weighing device |
| CN110954132A (en) * | 2019-10-31 | 2020-04-03 | 太原理工大学 | Method for carrying out navigation fault identification through GRNN (generalized regression neural network) assisted adaptive Kalman filtering |
| CN111259927A (en) * | 2020-01-08 | 2020-06-09 | 西北工业大学 | Rocket engine fault diagnosis method based on neural network and evidence theory |
| CN112651456A (en) * | 2020-12-31 | 2021-04-13 | 遵义师范学院 | Unmanned vehicle control method based on RBF neural network |
Non-Patent Citations (7)
| Title |
|---|
| Sharaf etc..Sensor Integration for Satellite-Based Vehicular Navigation Using Neural Networks.2007,第Vol.18卷(第No.2期),全文. * |
| 基于信息融合的汽车衡称重传感器故障诊断;林海军;滕召胜;迟海;吴阳平;易钊;;控制理论与应用(第01期);全文 * |
| 基于多传感器信息融合的汽车衡误差补偿;林海军;滕召胜;迟海;易钊;邬蓉蓉;;仪器仪表学报(第06期);全文 * |
| 孙勇 ; 李宝聚 ; 孙志博 ; 李振元 ; 张罗宾 ; .融合RBF神经网络和集对分析的风电功率超短期预测.昆明理工大学学报(自然科学版).2020,(第05期),全文. * |
| 林海军 ; 滕召胜 ; 迟海 ; 吴阳平 ; 易钊 ; .基于信息融合的汽车衡称重传感器故障诊断.控制理论与应用.2010,(第01期),全文. * |
| 林海军 ; 滕召胜 ; 迟海 ; 易钊 ; 邬蓉蓉 ; .基于多传感器信息融合的汽车衡误差补偿.仪器仪表学报.2009,(第06期),全文. * |
| 融合RBF神经网络和集对分析的风电功率超短期预测;孙勇;李宝聚;孙志博;李振元;张罗宾;;昆明理工大学学报(自然科学版)(第05期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114894289A (en) | 2022-08-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114894289B (en) | Large-mass comparator based on data fusion algorithm | |
| CN112903235B (en) | Multi-element calibration method for thrust balance capable of completely simulating test state | |
| CN110823537B (en) | Method for detecting and evaluating performance of wind vibration control damper of super high-rise building | |
| CN107741269B (en) | Weighing sensor test compensation method based on fuzzy recognition | |
| CN114676736B (en) | Method for predicting surface quality of rolled workpiece | |
| CN112816216B (en) | Rolling bearing performance test bench and identification and correction method of abnormal test sample | |
| CN116678489B (en) | Quality control method, system, equipment and medium for force sensor | |
| CN108846200B (en) | Quasi-static bridge influence line identification method based on iteration method | |
| CN114722332B (en) | Wind tunnel balance calibration data processing method | |
| CN110737958A (en) | method for identifying weak links of rigidity of elastic support beam by using curvature corner indexes | |
| CN107657073A (en) | A kind of parameter identification method of sandwich composite structure | |
| CN114594118A (en) | Aluminum alloy structure microsegregation quantitative characterization method based on microbeam X-ray fluorescence | |
| CN112816001B (en) | Jacking process-based synchronous testing method for load displacement of bearings of multiple support shaft systems | |
| CN116561646A (en) | Rolling bearing multidimensional parameter automatic detection line and rolling bearing grade evaluation method | |
| CN106526533A (en) | Microporous MEMS acoustic array sensor and use method thereof | |
| CN109933031A (en) | A kind of system and method automatically correcting soft measuring instrument according to analysis data | |
| CN113704922B (en) | Method for predicting surface roughness based on sound vibration and texture characteristics | |
| CN112747773B (en) | Method for improving precision of gyroscope based on Allan variance and random polynomial | |
| CN104101418A (en) | Electronic analytical balance trace loading and drift discrimination method | |
| CN111595433B (en) | Position determination method and system for vibration sensor of whole aircraft engine | |
| CN114084764A (en) | Elevator transportation quality detection method and detection system | |
| CN117094608B (en) | Titanium belt production control method and system combining application requirements | |
| CN117091801B (en) | Balance calibration method based on two-degree-of-freedom calibration equipment | |
| CN117190887B (en) | Aerogel thickness automatic detection system for mobile phone production | |
| US11879869B2 (en) | Method for predicting surface quality of burnishing workpiece |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |