Balancing structure of freezing positioning centrifugal device

Abstract

The invention provides a balancing structure of a freezing and positioning centrifugal device.The balancing structure of the freezing and positioning centrifugal device.The balancing structure of the freezing and positioning centrifugal device.The balancing structure of the freezing and positioning centrifugal device.The balancing structure of the freezing and positioning centrifugal device.The balancing structure comprises a cross-shaped centrifugal frame, four trays are movably connected to the cross-shaped centrifugal frame, the trays are the same in shape and internally provided with a plurality of pipe cavities, and the pipe cavities are used for containing serum pipes; the serum tube comprises a sample tube and a balancing tube, the bottom end of the cross centrifugal frame is fixedly connected with an output shaft of the motor, the other end of the output shaft extends to the vibration conduction bin, a pcb is arranged at the bottom end of the vibration conduction bin, a gyroscope and a control chip are arranged on the pcb, a balancing indication module is arranged in the control chip, and the balancing indication module is connected with the motor. The trimming indication module is used for judging whether samples in the plurality of serum tubes in each tray are trimmed or not according to data acquired by the gyroscope, and calculating to obtain coordinates of different serum tubes; according to the invention, the problem of balancing the interior of the refrigerated centrifuge can be solved.

Description

Technical Field

[0001] The invention belongs to the technical field of refrigerated centrifuges, and in particular relates to a balancing structure of a refrigerated positioning centrifuge device. Background Art

[0002] In the field of modern medical testing and scientific research, serum testing is a vital task. Accurate serum test results play an irreplaceable role in the diagnosis of diseases, the formulation of treatment plans, and the advancement of scientific research. In the serum testing process, centrifugation is a key link, and the frozen positioning centrifuge plays an important role in it.

[0003] During the centrifugation process, ensuring the stability of the centrifuge device is an important prerequisite for obtaining accurate test results. However, due to the diversity of serum samples and various factors during operation, the centrifuge device often faces the problem of weight imbalance during operation.

[0004] Traditional centrifuges often lack effective balancing structures and methods when performing serum centrifugation. When serum tubes are placed in the centrifuge, if the weight distribution between the tubes is uneven, it will cause centrifugal instability. This unbalanced state will cause the centrifuge to vibrate violently during high-speed rotation, which will not only affect the centrifugal effect, but may also cause serious wear and tear on the equipment, causing equipment downtime. Summary of the invention

[0005] In view of this, the present invention provides a balancing structure for a refrigerated positioning centrifuge device, which can solve the problem of balancing the interior of a refrigerated centrifuge.

[0006] The present invention is achieved in that:

[0007] The present invention provides a balancing structure of a freezing positioning centrifuge, which comprises a cross centrifugal frame, four groups of trays are movably connected on the cross centrifugal frame, the trays have the same shape and are provided with a plurality of cavities inside, the cavities are used to place serum tubes, the serum tubes include sample tubes and balancing tubes, the bottom end of the cross centrifugal frame is fixedly connected to the output shaft of a motor, the other end of the output shaft extends to a vibration conduction chamber, a PCB board is arranged at the bottom end of the vibration conduction chamber, a gyroscope and a control chip are arranged on the PCB board, a balancing indication module is arranged in the control chip, the balancing indication module is used to judge whether the samples in the plurality of serum tubes in each of the trays are balanced according to the data collected by the gyroscope, and calculate the coordinates of the difference serum tubes, the difference serum tubes refer to the serum tubes whose sample weight does not match the sample weight in the serum tubes at the corresponding positions in the other three trays.

[0008] On the basis of the above technical scheme, the beneficial effects of the balancing structure of a refrigerated positioning centrifuge device of the present invention are: using a gyroscope to collect the vibration value generated by the synchronous motor when the centrifuge frame is unbalanced, and cooperating with the balancing indication module in the control chip to find the location of the difference serum tube, and using the balancing tube to replace the difference serum tube for weight balancing, wherein the content of the balancing tube is water, and the balancing tube can also be replaced by other types of weights.

[0009] On the basis of the above technical solution, the balancing structure of a refrigerated positioning centrifuge device of the present invention can also be improved as follows:

[0010] In which, the tray is movably connected to the cross centrifugal frame through a centrifugal suspension, and a group of suspension openings are provided at the four ends of the tray, and the suspension openings are U-shaped, V-shaped or arc-shaped. A group of the suspension openings includes raised wings on both sides, and the raised wings are parallel to each other. A centrifugal shaft is provided between the raised wings on both sides through a through hole, and the centrifugal suspension is movably connected to the suspension opening through the centrifugal shaft. The centrifugal shaft is used to realize that when the motor drives the cross centrifugal frame to rotate, the centrifugal suspension and the tray inside are driven to deviate from the direction of the earth's gravity along the normal of the bottom wall of the tray with the centrifugal shaft as the axis, thereby achieving a centrifugal effect.

[0011] Furthermore, when all the tube cavities on each group of the tray are empty or fully loaded with serum tubes, the center of gravity of the tray and its geometric center of gravity are on the same vertical line, and the normal of the bottom wall of the tray remains parallel to the direction of the earth's gravity.

[0012] Furthermore, each group of the trays is provided with M cavities, where M=12-20.

[0013] Furthermore, the output shaft of the motor includes an upper shaft body portion and a lower shaft body portion, a shaft sleeve is provided at the center of the bottom wall of the cross centrifugal frame, the cross centrifugal frame is fixedly connected to the upper shaft body portion through the bottom shaft sleeve, and the lower shaft body portion is abutted against the inner wall of the vibration conduction bin, so as to transmit vibration to the gyroscope on the PCB board through the vibration conduction bin.

[0014] The beneficial effects of adopting the above-mentioned improvement scheme are as follows: by arranging a shaft sleeve to connect the cross centrifugal frame with the output shaft of the motor, the cross centrifugal frame is driven to rotate by the rotation of the motor, and the lower shaft body part abuts against the inner wall of the vibration conduction bin, so that when the weight of the serum tubes inside the tray on the cross centrifugal frame is unbalanced, the motor will vibrate due to the centrifugal action, and the vibration will be transmitted to the vibration conduction bin by the lower shaft body part, and then the vibration value will be collected by the gyroscope.

[0015] Furthermore, the tray is installed inside the centrifugal suspension, and hanging ears are provided on both sides of the centrifugal suspension. The hanging ears are provided with hanging holes, and the hanging holes are sleeved on the centrifugal shaft.

[0016] Furthermore, a centrifugal shaft is provided on the through hole of each group of the raised wings, one end of the centrifugal shaft is fixedly connected to the through hole, and the other end passes through the hanging hole, and the extension length of the other end passing through the hanging hole is less than 1 cm.

[0017] Furthermore, the M cavities are centrally symmetric with respect to the center of the tray.

[0018] Furthermore, it also includes a non-magnetic kit and a magnetic angle encoder. The non-magnetic kit is fixedly sleeved with the lower shaft body, and the magnetic angle encoder is arranged on the PCB board. The magnetic angle encoder is located directly below the non-magnetic kit.

[0019] The beneficial effect of adopting the above-mentioned improvement scheme is as follows: by providing a non-magnetic conductive kit to prevent the motor from being magnetically conductive through the output shaft to the bottom during the output process, the interference of the magnetic field on the gyroscope data acquisition is reduced.

[0020] Furthermore, the trim indication module is used to perform the following steps:

[0021] S10: changing the motor speed so that after the motor is stabilized at the first speed, a sequence of rotation parameters of the gyroscope is acquired in real time, recorded as a first speed parameter sequence, wherein the rotation parameters include angular velocity, angular acceleration and angular displacement;

[0022] S20: performing a fast Fourier transform on the first speed parameter sequence to obtain a corresponding frequency domain feature, recorded as a first frequency domain feature;

[0023] S30: changing the motor speed so that after the motor is stabilized at the second speed, a sequence of rotation parameters of the gyroscope is acquired in real time, which is recorded as a second speed parameter sequence;

[0024] S40: performing a fast Fourier transform on the second speed parameter sequence to obtain a corresponding frequency domain feature, recorded as a second frequency domain feature;

[0025] S50: According to the preset first standard feature and the second standard feature, the first frequency domain feature and the second frequency domain feature are respectively calculated to obtain a first difference feature and a second difference feature; wherein the first standard feature is the frequency domain feature obtained by using the steps S10-S20 under the pre-balancing condition and is recorded as the first standard feature; the second standard feature is the frequency domain feature obtained by using the steps S30-S40 under the pre-balancing condition and is recorded as the second standard feature;

[0026] S60: Inputting the obtained first difference feature and the second difference feature into a pre-trained difference calculation model to obtain a difference transformation vector;

[0027] S70: performing matrix multiplication operation on the difference transformation vector and a preset basic matrix to obtain an identification matrix, wherein the identification matrix is ​​a 4×M matrix, where M is the number of serum tubes on each tray; an element of 1 in the identification matrix represents a difference serum tube, and an element of 0 represents a normal serum tube;

[0028] S80: Outputting the position coordinates and weight difference estimation values ​​of the difference serum tubes in each tray according to the identification matrix.

[0029] Compared with the prior art, the balancing structure of a freezing positioning centrifugal device provided by the present invention has the following beneficial effects: after the centrifugal device is balanced so that the weight between each tray is uniform, the accuracy and reliability of the centrifugal effect can be greatly improved. In serum testing, accurate separation is the basis for obtaining accurate results. Uniform weight distribution ensures the consistency of centrifugal force in all directions, so that different components in serum can be separated in a predetermined manner. For example, when separating blood cells and serum, uniform centrifugal force can make blood cells quickly and completely precipitate to the bottom of the centrifuge tube, while serum is in the upper layer, thereby achieving a clear separation interface. This provides pure and accurate samples for subsequent detection and analysis, greatly improving the reliability of the test results; the centrifugal operation with uniform weight can also significantly reduce the wear of the driving components of the centrifugal device and extend the service life of the centrifugal device. Unbalanced centrifugation will cause the device to vibrate violently during high-speed rotation, and this vibration will cause great stress and friction to key components such as the motor, cross centrifugal frame, and tray. After balancing, the centrifugal device runs smoothly, and the stress and friction borne by each component are greatly reduced. The motor can work in a stable state, reducing problems such as winding loosening and bearing wear caused by vibration; the overall life of the centrifugal device is effectively extended, and the maintenance cost and component replacement frequency of the equipment are reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0031] Figure 1 It is a structural schematic diagram of a balancing structure of a freezing positioning centrifuge device;

[0032] Figure 2A schematic diagram of the interior of a sensor cavity of a balancing structure of a freezing positioning centrifugal device;

[0033] Figure 3 A schematic diagram of a centrifugal suspension of a balancing structure of a refrigerated positioning centrifugal device;

[0034] In the accompanying drawings, the components represented by the reference numerals are listed as follows:

[0035] 10. Cross centrifugal rack; 11. Tray; 12. Tube cavity; 13. Serum tube; 14. Motor; 15. Vibration conduction chamber; 16. PCB board; 17. Gyroscope; 18. Control chip; 19. Centrifugal suspension; 20. Suspension opening; 21. Centrifugal shaft; 22. Non-magnetic conductive kit. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0037] Embodiment 1:

[0038] like Figure 1 , as shown in FIG2 , is a first embodiment of a balancing structure of a freezing positioning centrifuge device provided by the present invention. In this embodiment, it includes a cross centrifugal frame 10, on which four groups of trays 11 are movably connected. The trays 11 have the same shape and are provided with multiple cavities 12 inside. The cavities 12 are used to place serum tubes 13. The serum tubes 13 include sample tubes and balancing tubes. The bottom end of the cross centrifugal frame 10 is fixedly connected to the output shaft of the motor 14, and the other end of the output shaft extends to the vibration conduction chamber 15. A PCB board 16 is provided at the bottom end of the vibration conduction chamber 15. A gyroscope 17 and a control chip 18 are provided on the PCB board 16. A balancing indication module is provided in the control chip 18. The balancing indication module is used to determine whether the samples in the multiple serum tubes 13 in each tray 11 are balanced according to the data collected by the gyroscope 17, and calculate the coordinates of the difference serum tubes. The difference serum tubes refer to the serum tubes 13 whose sample weights do not match the sample weights of the serum tubes 13 at the corresponding positions in the other three trays 11.

[0039] Among them, in the above technical solution, the balancing tube has the same size as the sample tube and acts as a balancing weight. Its weight can be controlled by adding different amounts of water. During the balancing process, the balancing tube can be clamped by a robotic arm and placed into the corresponding tube cavity 12.

[0040] Among them, in the above technical scheme, the tray 11 is movably connected to the cross centrifugal frame 10 through the centrifugal suspension 19, and a group of suspension openings 20 are provided at the four ends of the tray 11. The suspension openings 20 are U-shaped, V-shaped or arc-shaped. A group of suspension openings 20 include raised wings on both sides, and the raised wings are parallel to each other. A centrifugal shaft 21 is provided between the raised wings on both sides through a through hole. The centrifugal suspension 19 is movably connected to the suspension opening 20 through the centrifugal shaft 21. The centrifugal shaft 21 is used to realize that when the motor 14 drives the cross centrifugal frame 10 to rotate, the centrifugal suspension 19 and the tray 11 inside are driven to deviate from the direction of the earth's gravity along the normal of the bottom wall of the tray 11 with the centrifugal shaft 21 as the axis, thereby achieving a centrifugal effect.

[0041] Furthermore, in the above technical solution, when all the tube cavities 12 on each group of trays 11 are empty or full of serum tubes 13, the center of gravity of the tray 11 and its geometric center of gravity are on the same vertical line, and the normal of the bottom wall of the tray 11 remains parallel to the direction of the earth's gravity.

[0042] Furthermore, in the above technical solution, each group of trays 11 is provided with M cavities 12, where M=12-20.

[0043] Furthermore, in the above technical solution, the output shaft of the motor 14 includes an upper shaft body portion and a lower shaft body portion, a shaft sleeve is provided at the center of the bottom wall of the cross centrifugal frame 10, the cross centrifugal frame 10 is fixedly connected to the upper shaft body portion through the bottom shaft sleeve, and the lower shaft body portion is abutted against the inner wall of the vibration conduction chamber 15, and is used to transmit vibration to the gyroscope 17 on the PCB board 16 through the vibration conduction chamber 15.

[0044] Furthermore, in the above technical solution, the tray 11 is installed inside the centrifugal suspension 19 , and hanging ears are provided on both sides of the centrifugal suspension 19 . The hanging ears are provided with hanging holes, and the hanging holes are sleeved on the centrifugal shaft 21 .

[0045] Furthermore, in the above technical solution, a centrifugal shaft 21 is provided on the through hole of each group of raised wings, one end of the centrifugal shaft 21 is fixedly connected to the through hole, and the other end passes through the hanging hole, and the extension length of the other end passing through the hanging hole is less than 1 cm.

[0046] Furthermore, in the above technical solution, the M cavities 12 form central symmetry with respect to the center of the tray 11 .

[0047] Furthermore, the above technical solution also includes a non-magnetic kit 22 and a magnetic angle encoder. The non-magnetic kit 22 is fixedly sleeved with the lower shaft portion, and the magnetic angle encoder is arranged on the PCB board 16. The magnetic angle encoder is located directly below the non-magnetic kit 22.

[0048] Further, in the above technical solution, wherein, in the above technical solution, the balancing indication module is used to perform the following steps:

[0049] S10, changing the motor speed so that after the motor stabilizes at a first speed, a sequence of rotation parameters of the gyroscope is acquired in real time, recorded as a first speed parameter sequence, wherein the rotation parameters include angular velocity, angular acceleration and angular displacement;

[0050] S20, performing a fast Fourier transform on the first speed parameter sequence to obtain a corresponding frequency domain feature, recorded as a first frequency domain feature;

[0051] S30, changing the motor speed so that after the motor is stabilized at the second speed, a sequence of rotation parameters of the gyroscope is acquired in real time, which is recorded as a second speed parameter sequence;

[0052] S40, performing a fast Fourier transform on the second speed parameter sequence to obtain a corresponding frequency domain feature, recorded as a second frequency domain feature;

[0053] S50, according to the preset first standard feature and the second standard feature, respectively performing difference calculation on the first frequency domain feature and the second frequency domain feature to obtain a first difference feature and a second difference feature; wherein the first standard feature is the frequency domain feature obtained by using the steps S10-S20 under the pre-balancing condition and is recorded as the first standard feature; the second standard feature is the frequency domain feature obtained by using the steps S30-S40 under the pre-balancing condition and is recorded as the second standard feature;

[0054] S60, inputting the obtained first difference feature and the second difference feature into a pre-trained difference calculation model to obtain a difference transformation vector;

[0055] S70, performing a matrix multiplication operation on the difference transformation vector and a preset basic matrix to obtain an identification matrix, wherein the identification matrix is ​​a 4×M matrix, where M is the number of serum tubes on each tray; an element of 1 in the identification matrix represents a difference serum tube, and an element of 0 represents a normal serum tube;

[0056] S80. Outputting the position coordinates and weight difference estimation values ​​of the difference serum tubes in each tray according to the identification matrix.

[0057] The difference calculation model adopts a deep neural network structure, specifically a hybrid model including a multi-layer convolutional neural network and a long short-term memory network (LSTM); the convolution layer is used to extract the spatial correlation of frequency domain features, and the LSTM layer is used to capture the long-term dependency of time series.

[0058] The training steps of the difference calculation model are:

[0059] Step 1: Collect a large amount of balanced and unbalanced sample data;

[0060] Step 2: Preprocess and standardize sample data;

[0061] Step 3: Divide the processed data into training set and validation set;

[0062] Step 4: Use back propagation algorithm and stochastic gradient descent method to train the model;

[0063] Step 5: Optimize model performance by adjusting hyperparameters and network structure;

[0064] Step 6: Evaluate the model on the validation set until it reaches the predetermined accuracy requirement.

[0065] During the training of the difference calculation model, the step of establishing the training data set is to use an experimental method to traverse the different weight combinations of each serum tube sample on the tray.

[0066] The specific steps are as follows:

[0067] Step 1: Set a series of discrete sample weight values, such as 0g, 1g, 2g, ..., 10g...;

[0068] Step 2: For each serum tube position on each tray, try different weight values ​​in turn;

[0069] Step 3: For each weight combination, multiple centrifugation experiments were performed to collect gyroscope data;

[0070] Step 4: Perform fast Fourier transform on the collected data to obtain frequency domain features;

[0071] Step 5: Associate the frequency domain features with the corresponding weight distribution to form training data pairs;

[0072] Step 6: Repeat the above steps until all possible weight distribution combinations are covered;

[0073] Step 7: Clean and standardize the combined data;

[0074] Step 8: Divide the processed data set into training set and validation set for model training and evaluation.

[0075] The specific implementation methods of the above steps are described in detail below:

[0076] S10, changing the motor speed so that after the motor stabilizes at a first speed, a sequence of rotation parameters of the gyroscope is acquired in real time, recorded as a first speed parameter sequence, wherein the rotation parameters include angular velocity, angular acceleration and angular displacement.

[0077] The specific implementation is as follows:

[0078] Step 1: Motor speed control: Motor speed control can be achieved by adjusting the input voltage or current of the motor. Assuming a DC motor is used, its speed is approximately linearly related to the input voltage, which can be expressed as:

[0079] n=k n V-n0;

[0080] Where n is the motor speed (unit: revolutions per minute), k n is the speed constant of the motor (unit: (rpm) / volt), V is the input voltage (unit: volt), and n0 is the no-load speed of the motor (unit: rpm).

[0081] k n and n0 can be measured experimentally, the specific steps are as follows:

[0082] Step 1: Measure the motor speed at different input voltages;

[0083] Step 2: Draw the speed-voltage curve;

[0084] Step 3: Use the least squares method to fit the straight line and get k n and n0.

[0085] Step 2: Obtain gyroscope parameters: The gyroscope can measure angular velocity ω(t), angular acceleration α(t) and angular displacement θ(t) in real time. There is the following relationship between these parameters:

[0086]

[0087] Assuming the sampling interval is Δt, these relationships can be approximated in discrete time series as:

[0088]

[0089] θ[k]≈θ[k-1]+ω[k-1]Δt.

[0090] Where k is the index of the discrete time series.

[0091] Step 3: Parameter sequence representation: The first speed parameter sequence can be represented as:

[0092] S1={(ω1[k],α1[k],θ1[k])|k=0,1,...,N-1};

[0093] Where N is the number of sampling points, ω1[k], α1[k] and θ1[k] represent the angular velocity, angular acceleration and angular displacement at the first speed, respectively.

[0094] S20. Perform fast Fourier transform on the first rotational speed parameter sequence to obtain corresponding frequency domain features, recorded as first frequency domain features.

[0095] The specific implementation is as follows:

[0096] Step 1: Fast Fourier Transform (FFT): For each parameter sequence (angular velocity, angular acceleration, angular displacement), perform FFT separately. Take angular velocity as an example:

[0097]

[0098] Where Ω1[m] is the frequency domain representation of the angular velocity, m=0,1,...,N-1 is the frequency index, and j is the imaginary unit.

[0099] Similarly, perform FFT on angular acceleration and angular displacement:

[0100]

[0101] Step 2: Frequency domain feature extraction: The amplitude spectrum and phase spectrum of the frequency domain signal can be extracted as features. For angular velocity:

[0102] Amplitude spectrum:

[0103] Phase spectrum:

[0104] Similarly, the amplitude spectrum and phase spectrum of angular acceleration and angular displacement can be obtained.

[0105] Step 3: First frequency domain feature representation:

[0106] The first frequency domain feature can be expressed as:

[0107] F1={(|Ω1[m]|,∠Ω1[m],|A1[m]|,∠A1[m],|Θ1[m]|,∠Θ1[m])|m=0,1,...,N-1}.

[0108] S30, changing the motor speed so that after the motor is stabilized at the second speed, a sequence of rotation parameters of the gyroscope is acquired in real time, which is recorded as a second speed parameter sequence.

[0109] The specific implementation is similar to S10, except that the motor speed is adjusted to the second target value. It can be expressed by a similar formula:

[0110] n2=k n V2-n0;

[0111] Wherein, n2 is the second speed, and V2 is the corresponding input voltage.

[0112] The second speed parameter sequence can be expressed as:

[0113] S2={(ω2[k],α2[k],θ2[k])|k=0,1,...,N-1}.

[0114] S40. Perform fast Fourier transform on the second rotation speed parameter sequence to obtain corresponding frequency domain features, recorded as second frequency domain features.

[0115] The specific implementation is similar to S20, except that the input sequence is changed to S2. The second frequency domain feature finally obtained can be expressed as:

[0116] F2={(|Ω2[m]|,∠Ω2[m],|A2[m]|,∠A2[m],|Θ2[m]|,∠Θ2[m])|m=0,1,...,M-1}.

[0117] S50: Calculate the difference between the first frequency domain feature and the second frequency domain feature according to the preset first standard feature and the second standard feature to obtain the first difference feature and the second difference feature.

[0118] The specific implementation is as follows:

[0119] Step 1: Standard characteristics: Assume that the standard characteristics obtained under pre-balancing conditions are:

[0120] F 1std ={(|Ω 1std [m]|,∠Ω 1std [m],|A 1std [m]|,∠A 1std [m],|Θ 1std [m]|,∠Θ 1std [m])|m=0,1,...,N-1};

[0121] F 2std ={(|Ω 2std [m]|,∠Ω 2std [m],|A 2std [m]|,∠A 2std [m],|Θ 2std [m]|,∠Θ 2std [m])|m=0,1,...,N-1}.

[0122] Step 2: Difference calculation: For the first speed, the difference characteristic can be expressed as:

[0123] ΔF1={(Δ|Ω1[m]|,Δ∠Ω1[m],Δ|A1[m]|,Δ∠A1[m],Δ|Θ1[m]|,Δ∠Θ1[m])|m=0,1,...,N-1};

[0124] in,

[0125] Δ|Ω1[m]|=|Ω1[m]|-|Ω 1std [m]|;

[0126] Δ∠Ω1[m]=∠Ω1[m]-∠Ω 1std [m];

[0127] The differences of other parameters are calculated in the same way.

[0128] Similarly, for the second rotation speed, the calculation method of the difference characteristic ΔF2 is similar.

[0129] S60: Input the obtained first difference feature and the second difference feature into a pre-trained difference calculation model to obtain a difference transformation vector.

[0130] The specific implementation is as follows:

[0131] Step 1: Difference calculation model structure: Assume that a hybrid model containing a convolutional neural network (CNN) and a long short-term memory network (LSTM) is used. The input of the model is ΔF1 and ΔF2, and the output is the difference transformation vector v.

[0132] The model structure can be expressed as:

[0133] h1=CNN(ΔF1);

[0134] h2=CNN(ΔF2);

[0135] h = Concat(h1,h2);

[0136] s = LSTM(h);

[0137] v = FC(s);

[0138] In the formula, CNN represents convolutional neural network, Concat represents concatenation operation, LSTM represents long short-term memory network, and FC represents fully connected layer.

[0139] Step 2: CNN structure: Assume that CNN contains L convolutional layers, the operation of each layer can be expressed as:

[0140] x (l+1) =ReLU(Conv(x (l) ,W (l) )+b (l) );

[0141] In the formula, x (l) is the input of the lth layer, W (l) is the convolution kernel weight, b (l) is the bias term and ReLU is the activation function.

[0142] Step 3: LSTM structure: The core formula of LSTM is as follows:

[0143] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0144] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0145]

[0146] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0147] h t =o t *tanh(C t );

[0148] In the formula, f t 、i t , o t They are forget gate, input gate and output gate respectively, C t is the cell state, h t is the hidden state, σ is the sigmoid function, and * represents element-by-element multiplication.

[0149] Step 4: Fully connected layer:

[0150] The operation of the fully connected layer can be expressed as:

[0151] v=W fc ·s+b fc ;

[0152] Where W fc is the weight matrix, b fc is the bias vector.

[0153] S70, performing matrix multiplication operation on the difference transformation vector and a preset basic matrix to obtain an identification matrix, wherein the identification matrix is ​​a 4×M matrix, where M is the number of serum tubes on each tray.

[0154] The specific implementation is as follows:

[0155] Step 1: Difference transformation vector: Assume that the dimension of the difference transformation vector v is d, that is

[0156] Step 2: Basic Matrix: The default basic matrix B has a dimension of d×(4M), that is,

[0157] Step 3: Matrix multiplication: The calculation process of the identification matrix I is as follows:

[0158] I′=v T B;

[0159] In the formula,

[0160] Step 4: Reshape the identity matrix: Reshape I′ into a 4×M matrix I:

[0161] I = reshape(I′,4,M);

[0162] Step 5: Threshold processing: In order to obtain a binary identification matrix, I needs to be thresholded:

[0163]

[0164] Where τ is the preset threshold, i=1,2,3,4, j=1,2,...,M.

[0165] S80. Outputting the position coordinates and weight difference estimation values ​​of the difference serum tubes in each tray according to the identification matrix.

[0166] The specific implementation is as follows:

[0167] Step 1: Position coordinate determination: For each element I in the identity matrix I ij , if I ij =1, it means that there is a difference in the position of the jth serum tube in the i-th tray. A coordinate mapping function g can be defined:

[0168] f(i,j)=(x ij ,y ij );

[0169] In the formula, (x ij ,y ij ) is the two-dimensional coordinate of the jth serum tube on the i-th tray. This mapping function can be defined according to the geometric layout of the actual tray.

[0170] Step 2: Weight difference estimation: In order to estimate the weight difference, a weight difference estimation model can be established. Assume that this model is a neural network, whose input is the difference transformation vector v and the position coordinate (x, y), and the output is the weight difference estimation value Δw.

[0171] The model can be expressed as:

[0172] Δw=g(v,x,y;θ);

[0173] In the formula, g represents the neural network function and θ is the model parameter.

[0174] The specific neural network structure can be designed as follows:

[0175] h1=ReLU(W1·v+b1);

[0176] h2=ReLU(W2·[h1,x,y]+b2);

[0177] Δw=W3·h2+b3;

[0178] Where W1, W2, W3 are weight matrices, b1, b2, b3 are bias vectors, and ReLU is the activation function.

[0179] Step 3: Output results: For each I ij =1, the following information can be output:

[0180] Tray number: i Serum tube position: (x ij ,y ij ) = f(i,j) weight difference estimate: Δw ij =g(v,x ij ,y ij ;θ)

[0181] The following is a specific embodiment, which describes a specific implementation scenario of the present invention in practical applications:

[0182] Embodiment 2:

[0183] Application of the balancing structure of the refrigerated positioning centrifuge in medical laboratories

[0184] A medical laboratory needs to centrifuge a large number of blood samples to separate serum and blood cells. In order to improve the centrifugal efficiency and ensure the accuracy of sample processing, the laboratory introduced a new type of freezing positioning centrifuge, which is equipped with the balancing structure described in the present invention.

[0185] 1. Basic configuration of centrifugal device

[0186] Each tray has 16 tube cavities, and a total of 64 serum tube samples can be processed simultaneously. The size of each tray is 20cm×20cm, and the tube cavities are arranged in a 4×4 matrix. The serum tubes used are standard 15mL centrifuge tubes with an outer diameter of 16mm and a height of 120mm.

[0187] The main components of the device include:

[0188] Motor: Maximum speed 6000rpm, rated power 500W

[0189] Gyroscope: sampling rate 1000Hz, angular velocity measurement range ±2000° / s, angular acceleration measurement range ±16g

[0190] Control chip: 32-bit ARM Cortex-M7 processor, main frequency 480MHz, built-in 2MB Flash and 1MB RAM

[0191] 2. Implementation of the balancing structure

[0192] 2.1 Motor speed control

[0193] The speed constant k of the motor is measured experimentally n =100rpm / V, no-load speed n0 = 100rpm. According to the formula n = k n V-n0, we can get the relationship between motor speed and input voltage.

[0194] For trim testing, two different speeds were selected for testing:

[0195] First speed: 1000rpm, corresponding to input voltage 11V

[0196] Second speed: 2000rpm, corresponding to input voltage 21V

[0197] 2.2 Gyroscope parameter collection

[0198] The gyroscope sampling interval Δt = 1ms. At each rotation speed, 10 seconds of data are collected, that is, 10,000 sampling points.

[0199] 2.3 Fast Fourier Transform

[0200] The FFTW library is used for fast Fourier transform. For each rotation speed, FFT is performed on the angular velocity, angular acceleration, and angular displacement sequences to obtain the corresponding frequency domain features.

[0201] 2.4 Difference calculation model

[0202] A hybrid model including CNN and LSTM is used as the difference calculation model. The model structure is as follows:

[0203] CNN part: 3 convolutional layers, each with a convolution kernel size of 3×3 and 32, 64, and 128 channels respectively

[0204] LSTM part: 2 layers of LSTM, the number of hidden units in each layer is 256

[0205] Fully connected layer: 2 layers, with 128 and 64 units respectively

[0206] The input of the model is the difference feature at two speeds, and the output is a 64-dimensional difference transformation vector.

[0207] 2.5 Basic Matrix

[0208] The default basic matrix B has a dimension of 64×64, is initialized to a random normally distributed value, and then optimized through training.

[0209] 2.6 Weight Difference Estimation Model

[0210] A 3-layer fully connected neural network is used, with 128 and 64 hidden layer units respectively and 1 unit in the output layer.

[0211] 3. Experimental Procedure

[0212] 3.1 Sample preparation

[0213] Prepare 64 15 mL centrifuge tubes, of which 63 tubes are filled with 10 mL of blood sample (density about 1.06 g / mL) and 1 tube is filled with 9.5 mL of blood sample to simulate an unbalanced situation.

[0214] The sample weight distribution is shown in the following table (unit: g):

[0215] Pallet No. Position 1 Position 2 Position 3 Position 4 … Position 16 1 10.60 10.60 10.60 10.60 … 10.60 2 10.60 10.60 10.60 10.60 … 10.60 3 10.60 10.60 10.60 10.60 … 10.60 4 10.07 10.60 10.60 10.60 … 10.60

[0216] Note: Position 1 of tray 4 is an unbalanced sample.

[0217] 3.2 Balancing test process

[0218] Step 1: Start the centrifugal device and adjust the motor speed to a first speed of 1000 rpm.

[0219] Step 2: Wait 5 seconds for the system to stabilize, then start collecting gyroscope data for 10 seconds.

[0220] Step 3: Perform fast Fourier transform on the collected data to obtain the first frequency domain feature.

[0221] Step 4: Adjust the motor speed to the second speed of 2000 rpm.

[0222] Step 5: Repeat steps 2 and 3 to obtain the second frequency domain feature.

[0223] Step 6: Calculate the difference between the first frequency domain feature and the second frequency domain feature and the preset standard feature.

[0224] Step 7: Input the difference feature into the difference calculation model to obtain the difference transformation vector.

[0225] Step 8: Multiply the difference transformation vector by the basis matrix to obtain the identity matrix.

[0226] Step 9: Determine the position coordinates of the differential serum tubes based on the identification matrix.

[0227] Step 10: Using the weight difference estimation model, calculate the weight difference estimates for the differential serum tubes.

[0228] 3.3 Experimental Results

[0229] After the above steps, the system successfully detected the unbalanced serum tube position and gave an estimate of the weight difference. The specific results are as follows:

[0230] Detected difference serum tube information: - Tray number: 4 - Serum tube position coordinates: (2.5cm, 2.5cm) - Weight difference estimate: -0.52g

[0231] The identification matrix I output by the system (partial display):

[0232] Pallet No. Position 1 Position 2 Position 3 Position 4 1 0 0 0 0 2 0 0 0 0 3 0 0 0 0 4 1 0 0 0

[0233] 4. Performance Analysis

[0234] 4.1 Detection Accuracy

[0235] The experiment was repeated 100 times, and each time a serum tube position was randomly selected as an unbalanced sample. The statistical results are as follows:

[0236] Number of correct detections: 98

[0237] Missed detection times: 1

[0238] Number of false detections: 1

[0239] Detection accuracy: 98%

[0240] 4.2 Weight difference estimation error

[0241] For the 98 samples that were correctly detected, the root mean square error (RMSE) between the estimated and actual weight differences was calculated:

[0242] RMSE = 0.038 g

[0243] This shows that the weight difference estimation has high accuracy, with an average error of around 0.04g.

[0244] 4.3 Detection time

[0245] The time consumed for each balancing test was recorded, and the statistical results are as follows:

[0246] Average detection time: 25.6 seconds

[0247] Shortest detection time: 24.8 seconds

[0248] Maximum detection time: 26.5 seconds

[0249] Among them, the average time consumption distribution of each step is:

[0250] Data collection: 20 seconds (10 seconds for each of the two speeds)

[0251] Fast Fourier transform: 0.8 seconds

[0252] Model calculation and result output: 4.8 seconds

[0253] Embodiment three:

[0254] Each tray is provided with 12 tube chambers for placing serum tubes. The device also includes components such as gyroscopes, motors and control chips.

[0255] 1. Device Overview

[0256] The main parameters of the centrifuge device are as follows: - Number of trays: 4 - Number of serum tubes per tray (M): 12 - Maximum centrifugal force: 12,000 × g - Speed ​​range: 100-15,000 rpm - Temperature control range: -20°C to 40°C

[0257] The layout of each tray is shown in the following table:

[0258] Location 1 2 3 4 A ○ ○ ○ ○ B ○ ○ ○ ○ C ○ ○ ○ ○

[0259] Among them, ○ represents the lumen position.

[0260] 2. Balancing test process

[0261] In this embodiment, an actual balancing test process is simulated. It is assumed that the experimenter has placed samples of different weights on four trays and now needs to perform a balancing test.

[0262] Step 1: Initialization

[0263] The control chip starts the trim indication module and prepares for trim detection.

[0264] Step 2: First speed data collection (S10)

[0265] The control chip instructs the motor to accelerate to the first speed of 3,000rpm. After the speed stabilizes, the gyroscope starts to collect data. Assume that the sampling frequency is 1,000Hz and the sampling time is 5 seconds. At this time, the first speed parameter sequence of 5,000 data points is obtained.

[0266] Step 3: First speed data processing (S20)

[0267] Perform fast Fourier transform (FFT) on the first speed parameter sequence to obtain the first frequency domain feature. To simplify the description, only some frequency domain features of angular velocity are listed:

[0268] Frequency (Hz) Amplitude Phase (radians) 0 314.16 0 50 0.52 1.23 100 0.31 -0.78 150 0.18 2.01 200 0.09 -1.56

[0269] Step 4: Second speed data collection (S30)

[0270] The control chip instructs the motor to accelerate to the second speed of 6,000 rpm. Similarly, 5 seconds of data are collected to obtain the second speed parameter sequence.

[0271] Step 5: Second speed data processing (S40)

[0272] Perform FFT on the second speed parameter sequence to obtain the second frequency domain feature. Some angular velocity frequency domain features are as follows:

[0273]

[0274]

[0275] Step 6: Difference calculation (S50)

[0276] Assume that the standard characteristics obtained under pre-balancing conditions are as follows: The first standard characteristic (part):

[0277] Frequency (Hz) Amplitude Phase (radians) 0 314.16 0 50 0.50 1.20 100 0.30 -0.75 150 0.17 2.00 200 0.08 -1.55

[0278] Second standard feature (part):

[0279] Frequency (Hz) Amplitude Phase (radians) 0 628.32 0 100 0.80 1.55 200 0.43 -0.90 300 0.25 2.30 400 0.11 -1.80

[0280] Calculate the difference features:

[0281] First difference feature (part):

[0282] Frequency (Hz) Amplitude difference Phase difference (radians) 0 0 0 50 0.02 0.03 100 0.01 -0.03 150 0.01 0.01 200 0.01 -0.01

[0283] Second difference feature (part):

[0284] Frequency (Hz) Amplitude difference Phase difference (radians) 0 0 0 100 0.03 0.02 200 0.02 -0.02 300 0.01 0.04 400 0.01 -0.01

[0285] Step 7: Difference calculation model (S60)

[0286] The first difference feature and the second difference feature are input into a pre-trained difference calculation model, which includes a 3-layer convolutional neural network and a 2-layer LSTM network.

[0287] Convolutional neural network structure: - Input layer: difference features - First convolution layer: 16 3x3 convolution kernels, stride = 1, padding = same - Second convolution layer: 32 3x3 convolution kernels, stride = 1, padding = same - Third convolution layer: 64 3x3 convolution kernels, stride = 1, padding = same - Fully connected layer: 128 neurons

[0288] LSTM network structure: - Input layer: CNN output - First LSTM layer: 64 units - Second LSTM layer: 32 units - Fully connected layer: 48 neurons (corresponding to the dimension of the difference transformation vector)

[0289] The model outputs a 48-dimensional difference transformation vector v. Assume that the obtained vector is:

[0290] v=[0.12,-0.05,0.03,0.18,-0.09,0.07,…,0.02]

[0291] Step 8: Calculation of identification matrix (S70)

[0292] Assume that the preset basic matrix B is a 48x48 identity matrix. Perform matrix multiplication:

[0293] I'=v^TB=v^T

[0294] Reshape I' into a 4x12 matrix I:

[0295]

[0296] Assuming the threshold τ = 0.10, perform threshold processing to obtain the binary identification matrix:

[0297]

[0298]

[0299] Step 9: Result output (S80)

[0300] Based on the identification matrix, the positions of the differential serum tubes can be determined. Assume that the coordinate system of the tray is as follows:

[0301] x-axis: 1-4 (from left to right)

[0302] y-axis: AC (from top to bottom)

[0303] The coordinate mapping function f(i,j) is defined as follows:

[0304] x=(j-1)%4+1

[0305] y=chr(ord('A')+(j-1) / / 4)

[0306] Where i represents the tray number and j represents the serum tube position number.

[0307] The weight difference estimation model uses a three-layer fully connected neural network:

[0308] Input layer: 50 neurons (48-dimensional difference transformation vector + 2-dimensional coordinates)

[0309] Hidden layer 1: 32 neurons, ReLU activation

[0310] Hidden layer 2: 16 neurons, ReLU activation

[0311] Output layer: 1 neuron, linear activation

[0312] Estimating the model based on the identity matrix and weight differences yields the following results:

[0313] 1. Pallet 1: - Position A1 (1, A), weight difference: +0.8g - Position A4 (4, A), weight difference: +1.2g - Position B3 (3, B), weight difference: -0.9g

[0314] 2. Pallet 2: - Position A4 (4, A), weight difference: +1.0g - Position C1 (1, C), weight difference: -0.7g

[0315] 3. Pallet 3: - Position A3 (3, A), weight difference: +0.9g - Position C1 (1, C), weight difference: +1.1g - Position C4 (4, C), weight difference: -0.8g

[0316] 4. Tray 4: - Position A2 (2, A), weight difference: -0.9g - Position B3 (3, B), weight difference: +1.0g

[0317] Based on these results, the experimenter can adjust the weight of the serum tubes at the corresponding positions to achieve a better balancing effect.

[0318] 3. Balancing adjustment

[0319] The experimenters made balance adjustments based on the above results:

[0320] 1. Tray 1: - Reduce 0.8g sample at position A1 - Reduce 1.2g sample at position A4 - Increase 0.9g sample at position B3

[0321] 2. Tray 2: - Reduce 1.0g sample at A4 position - Increase 0.7g sample at C1 position

[0322] 3. Tray 3: - Reduce 0.9g sample at A3 - Reduce 1.1g sample at C1 - Increase 0.8g sample at C4

[0323] 4. Tray 4: -Add 0.9g sample at position A2 -Subtract 1.0g sample at position B3

[0324] 4. Test again

[0325] After the adjustment is completed, the experimenter restarts the balancing test program. After the same steps (S10-S80), a new identification matrix is ​​obtained:

[0326] 1 2 3 4 5 6 7 8 9 10 11 12 Tray 1 0 0 0 0 0 0 0 0 0 0 0 0 Tray 2 0 0 0 0 0 0 0 0 0 0 0 0 Tray 3 0 0 0 0 0 0 0 0 0 0 0 0 Tray 4 0 0 0 0 0 0 0 0 0 0 0 0

[0327] The new identification matrix shows that all positions are 0, indicating that the balancing adjustment is successful and all pallets have reached a good balance state.

[0328] Embodiment 3: The device is equipped with the balancing structure of the present invention, including four trays arranged in a cross, a gyroscope, a motor and a control chip. Each tray is provided with 20 cavities for placing serum tubes.

[0329] 1. Device initialization

[0330] The lab technician first initializes the device. He turns on the device and the control chip starts running the trim indicator module. The system automatically performs the following steps:

[0331] 1.1 Motor speed control

[0332] According to the preset parameters, the system sets the motor speed to two different values: - First speed: 3000rpm - Second speed: 4500rpm

[0333] The system controls the speed by adjusting the motor input voltage. After experimental determination, the speed constant k of the motor n =100rpm / V, no-load speed n0 = 500rpm. Therefore, the corresponding input voltages are:

[0334] The voltage corresponding to the first speed: V1 = (3000 + 500) / 100 = 35V

[0335] The second speed corresponds to voltage: V2 = (4500 + 500) / 100 = 50V

[0336] 1.2 Gyroscope parameter settings

[0337] The system sets the sampling frequency of the gyroscope to 1000 Hz, that is, the sampling interval Δt = 0.001 s. Each measurement lasts for 5 seconds, so the number of sampling points N = 5000.

[0338] 2. Sample loading

[0339] According to the experimental plan, the experimenters put 80 serum tubes into four trays. Each tray contains 20 serum tubes, numbered 1 to 20. The serum tubes contain samples of different weights, ranging from 1g to 10g.

[0340] 3. Balancing test process

[0341] 3.1 First speed data collection

[0342] The system adjusts the motor speed to 3000 rpm, waits for 10 seconds for the speed to stabilize, and then starts collecting gyroscope data to obtain the first speed parameter sequence S1.

[0343] 3.2 First speed data processing

[0344] The system performs fast Fourier transform (FFT) on S1 to obtain the first frequency domain feature F1.

[0345] 3.3 Second speed data collection

[0346] The system adjusts the motor speed to 4500 rpm, and waits for 10 seconds for the speed to stabilize. Then, the gyroscope data is collected to obtain the second speed parameter sequence S2.

[0347] 3.4 Second speed data processing

[0348] The system performs FFT on S2 to obtain the second frequency domain feature F2.

[0349] 3.5 Difference calculation

[0350] The system compares F1 and F2 with the preset standard feature F 1std and F 2std The difference calculation is performed to obtain the difference characteristics ΔF1 and ΔF2.

[0351] 3.6 Difference calculation model processing

[0352] The system inputs ΔF1 and ΔF2 into a pre-trained difference calculation model. The model contains a 3-layer convolutional neural network and a 2-layer LSTM network. The model outputs a 32-dimensional difference transformation vector v.

[0353] 3.7 Identity Matrix Generation

[0354] The system performs matrix multiplication on the difference transformation vector v and the preset 80×32 basic matrix B to obtain a 1×80 vector. The system reshapes the vector into a 4×20 matrix and performs threshold processing (the threshold is set to 0.5) to obtain a binary identification matrix I.

[0355] 3.8 Result Output

[0356] Based on the identification matrix I, the system determines the position coordinates and weight difference estimates of the difference serum tubes in each tray.

[0357] 4. Test results

[0358] After processing, the system outputs the following results:

[0359] Pallet No. Serum tube position Coordinates (x,y) Estimated weight difference (g) 1 7 (3.5,2.0) +2.3 2 13 (6.5,3.5) -1.8 3 4 (2.0,1.5) +3.1 3 19 (9.5,4.5) -1.5 4 9 (4.5,2.5) +2.7

[0360] 5. Result analysis and processing

[0361] The experimenter checked the results output by the system and found that the weights of five serum tubes were significantly different from those of other serum tubes in the corresponding positions. He decided to check these serum tubes one by one:

[0362] 5.1 Tray 1, serum tube 7: - Planned weight: 5.0 g - Actual weight: 7.3 g - Processing: Remove excess 2.3 g sample

[0363] 5.2 Tray 2, serum tube 13: - Original weight: 6.0g - Actual weight: 4.2g - Processing: Supplement 1.8g sample

[0364] 5.3 Tray 3, serum tube 4: - Planned weight: 3.0 g - Actual weight: 6.1 g - Processing: Remove excess 3.1 g sample

[0365] 5.4 Tray 3, serum tube 19: - Original weight: 8.0 g - Actual weight: 6.5 g - Processing: Supplement 1.5 g sample

[0366] 5.5 Tray 4, serum tube 9: - Planned weight: 4.0 g - Actual weight: 6.7 g - Processing: Remove excess 2.7 g sample

[0367] 6. Retest

[0368] After adjusting the sample weights of the five serum tubes, the experimenter restarted the balancing test process. This time, the system did not detect any significant weight differences, indicating that all serum tubes had reached a balanced state.

[0369] 7. Centrifugation process

[0370] After confirming the balance, the experimenter started the centrifugation process. The centrifuge ran at 6000rpm for 30 minutes. During the whole process, the equipment remained stable without obvious vibration or abnormal sound.

[0371] 8. Result Verification

[0372] After the centrifugation was completed, the experimenter took out the samples for analysis and found that all samples had been evenly centrifuged, with no sample loss or inadequate separation due to imbalance.

[0373] 9. Data Recording and Analysis

[0374] The experimenters recorded the data of the entire process, including:

[0375] 9.1 Initial weight distribution:

[0376]

[0377] Note: The values ​​with * indicate the detected differential serum tubes.

[0378] 9.2 Adjusted weight distribution:

[0379]

[0380] Embodiment 4:

[0381] Example of steps to establish a training data set during the training of the differential calculation model: There are 20 serum tube positions on each tray. In order to ensure balance and stability during centrifugation, the team decided to implement an intelligent balancing system. The core of this system is a differential calculation model that is used to accurately identify unbalanced serum tube positions. In order to train this model, the team needs to establish a comprehensive training data set.

[0382] Step 1: Set the sample weight value

[0383] The R&D team first determined a series of discrete and equally spaced sample weight values. Considering the situations that may be encountered in actual applications, they selected 11 weight values ​​between 0g and 10g, with an interval of 1g. The details are as follows:

[0384]

[0385] Step 2: Design a weight combination

[0386] To cover all possible weight distribution combinations, the team devised a systematic approach to composition. They decided to start from a completely balanced state and then gradually introduce imbalance until a predetermined maximum imbalance was reached.

[0387] First, they defined a baseline configuration where all serum tubes were loaded with 5 g of sample (weight number 6). Then, they designed the following unbalanced situations:

[0388] 1. Single-point imbalance: change the weight at one position on one pallet 2. Double-point imbalance: change the weight at one position on two different pallets 3. Three-point imbalance: change the weight at one position on three different pallets 4. Four-point imbalance: change the weight at one position on four pallets

[0389] For each imbalance condition, they considered different degrees of weight difference, ranging from 1g to 5g.

[0390] Step 3: Generate Weight Distribution Combinations

[0391] Based on the above design, the team began to generate specific weight distribution combinations. Here are some example combinations:

[0392] 1. Fully balanced configuration: All positions of all trays: 5g

[0393] 2. Single point unbalanced configuration (taking tray 1 as an example): Position 1 of tray 1: 6g (+1g) All other positions: 5g

[0394] Position 1 of tray 1: 7g (+2g) All other positions: 5g

[0395] Position 1 of tray 1: 8g (+3g) All other positions: 5g

[0396] Position 1 of tray 1: 9g (+4g) All other positions: 5g

[0397] Position 1 of tray 1: 10g (+5g) All other positions: 5g

[0398] 3. Dual-point unbalanced configuration (taking tray 1 and tray 2 as an example): Position 1 of tray 1: 6g (+1g) Position 1 of tray 2: 4g (-1g) All other positions: 5g

[0399] Position 1 of tray 1: 7g (+2g) Position 1 of tray 2: 3g (-2g) All other positions: 5g

[0400] Position 1 of tray 1: 8g (+3g) Position 1 of tray 2: 2g (-3g) All other positions: 5g

[0401] Position 1 of tray 1: 9g (+4g) Position 1 of tray 2: 1g (-4g) All other positions: 5g

[0402] Position 1 of tray 1: 10g (+5g) Position 1 of tray 2: 0g (-5g) All other positions: 5g

[0403] 4. Three-point unbalanced configuration (taking pallets 1, 2 and 3 as an example): Position 1 of pallet 1: 6g (+1g) Position 1 of pallet 2: 4g (-1g) Position 1 of pallet 3: 6g (+1g) All other positions: 5g

[0404] Position 1 of tray 1: 7g (+2g) Position 1 of tray 2: 3g (-2g) Position 1 of tray 3: 7g (+2g) All other positions: 5g

[0405] Position 1 of tray 1: 8g (+3g) Position 1 of tray 2: 2g (-3g) Position 1 of tray 3: 8g (+3g) All other positions: 5g

[0406] Position 1 of tray 1: 9g (+4g) Position 1 of tray 2: 1g (-4g) Position 1 of tray 3: 9g (+4g) All other positions: 5g

[0407] Position 1 of tray 1: 10g (+5g) Position 1 of tray 2: 0g (-5g) Position 1 of tray 3: 10g (+5g) All other positions: 5g

[0408] 5. Four-point unbalance configuration: Position 1 of tray 1: 6g (+1g) Position 1 of tray 2: 4g (-1g) Position 1 of tray 3: 6g (+1g) Position 1 of tray 4: 4g (-1g) All other positions: 5g

[0409] Position 1 of tray 1: 7g (+2g) Position 1 of tray 2: 3g (-2g) Position 1 of tray 3: 7g (+2g) Position 1 of tray 4: 3g (-2g) All other positions: 5g

[0410] Position 1 of tray 1: 8g (+3g) Position 1 of tray 2: 2g (-3g) Position 1 of tray 3: 8g (+3g) Position 1 of tray 4: 2g (-3g) All other positions: 5g

[0411] Position 1 of tray 1: 9g (+4g) Position 1 of tray 2: 1g (-4g) Position 1 of tray 3: 9g (+4g) Position 1 of tray 4: 1g (-4g) All other positions: 5g

[0412] Position 1 of tray 1: 10g (+5g) Position 1 of tray 2: 0g (-5g) Position 1 of tray 3: 10g (+5g) Position 1 of tray 4: 0g (-5g) All other positions: 5g

[0413] Step 4: Perform centrifugation experiment and collect data

[0414] For each weight combination, the team conducted multiple centrifugation experiments, each lasting 5 minutes, during which gyroscope data was collected once a second. To ensure the reliability and representativeness of the data, each combination was repeated 5 times.

[0415] The following is an example of gyroscope data from a single experiment (only part of the data is shown):

[0416] Time(s) Angular velocity (rad / s) Angular acceleration (rad / s^2) Angular displacement (rad) 0 0.000 0.000 0.000 1 10.472 10.472 5.236 2 20.944 10.472 20.944 3 31.416 10.472 47.124 … … … … 298 314.159 0.000 46806.784 299 314.159 0.000 47120.943 300 314.159 0.000 47435.102

[0417] Step 5: Data preprocessing and feature extraction

[0418] After collecting the raw data, the team preprocessed and extracted features. First, they performed a fast Fourier transform (FFT) on each parameter sequence (angular velocity, angular acceleration, angular displacement) to obtain frequency domain features. Then, they extracted the amplitude spectrum and phase spectrum of each frequency domain signal as features.

[0419] The following is an example of frequency domain features after FFT processing (only part of the data is shown):

[0420]

[0421] Step 6: Build training data pairs

[0422] The team associated the extracted frequency domain features with the corresponding weight distribution to form training data pairs. Each training data pair contains the following information:

[0423] 1. Input features: frequency domain features (amplitude spectrum and phase spectrum of angular velocity, angular acceleration, angular displacement) 2. Output labels: weight values ​​of 20 locations on each pallet (80 values ​​in total)

[0424] The following is an example of a training data pair (only part of the data is shown):

[0425] Input features: [314.159, 0.000, 0.001, 1.571, …, 0.000, 0.785]

[0426] Output labels: [5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5, #Tray 1 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5, #Tray 2 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5, #Tray 3 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5] #Tray 4

[0427] Step 7: Expand the dataset

[0428] In order to increase the diversity of the dataset and the generalization ability of the model, the team also considered the following situations:

[0429] 1. Multi-point imbalance: randomly select 1-3 locations on each pallet for weight adjustment 2. Different degrees of imbalance: in one experiment, different degrees of weight changes are applied to different locations 3. Random background noise: simulate the tiny vibrations and noise that may exist in the actual environment

[0430] These additional cases significantly increase the size and complexity of the dataset. The team estimates that the final training dataset will contain about 100,000 training examples.

[0431] Step 8: Data cleaning and standardization

[0432] After constructing the complete dataset, the team cleaned and standardized the data:

[0433] 1. Remove outliers: Use the box plot method to identify and remove possible abnormal data points 2. Standardization: Perform z-score standardization on each feature so that its mean is 0 and standard deviation is 1

[0434] Example of normalized features:

[0435] feature Original value Normalized value Feature 1 314.159 2.718 Feature 2 0.001 -0.532 Feature 3 1.571 0.867 … … … Features 1800 0.785 0.123

[0436] Step 9: Dataset Partitioning

[0437] Finally, the team divided the processed data set into training set, validation set and test set in a ratio of 7:2:1. The specific numbers are as follows:

[0438] Training set: 70,000 samples

[0439] Validation set: 20,000 samples

[0440] Test set: 10,000 samples

[0441] This division ensures that there is enough data for model training while retaining enough samples for validation and final performance evaluation.

Claims

1. A balancing structure for a refrigerated positioning centrifuge, characterized in that: The invention comprises a cross centrifugal frame (10), wherein four groups of trays (11) are movably connected to the cross centrifugal frame (10), wherein the trays (11) have the same shape and are provided with a plurality of cavities (12) therein, wherein the cavities (12) are used to place serum tubes (13), wherein the serum tubes (13) include sample tubes and balancing tubes, wherein the bottom end of the cross centrifugal frame (10) is fixedly connected to the output shaft of a motor (14), wherein the other end of the output shaft extends to a vibration conduction chamber (15), wherein a PCB (16) is provided at the bottom end of the vibration conduction chamber (15), wherein the PCB A gyroscope (17) and a control chip (18) are provided on the plate (16); a balancing indication module is provided in the control chip (18); the balancing indication module is used to determine whether the samples in the plurality of serum tubes (13) in each tray (11) are balanced according to the data collected by the gyroscope (17), and to calculate the coordinates of a differential serum tube, wherein the differential serum tube refers to a serum tube (13) whose sample weight does not match the sample weight of serum tubes (13) at corresponding positions in the other three trays (11).

2. The balancing structure of a refrigerated positioning centrifuge according to claim 1, characterized in that: The tray (11) is movably connected to the cross centrifugal frame (10) through a centrifugal suspension (19); a group of suspension openings (20) are provided at the four ends of the tray (11); the suspension openings (20) are U-shaped, V-shaped or arc-shaped; a group of the suspension openings (20) includes raised wings on two sides, the raised wings are parallel to each other, a centrifugal shaft (21) is provided between the raised wings on both sides through a through hole; the centrifugal suspension (19) is movably connected to the suspension opening (20) through the centrifugal shaft (21); the centrifugal shaft (21) is used to realize that when the motor (14) drives the cross centrifugal frame (10) to rotate, the centrifugal suspension (19) and the tray (11) inside are driven to deviate from the direction of the earth's gravity along the normal of the bottom wall of the tray (11) with the centrifugal shaft (21) as the axis, so as to achieve a centrifugal effect.

3. The balancing structure of a refrigerated positioning centrifuge according to claim 2, characterized in that: When all the tube cavities (12) on each group of the trays (11) are empty or fully loaded with the serum tubes (13), the center of gravity of the trays (11) and their geometric center of gravity are on the same vertical line, and the normal of the bottom wall of the trays (11) remains parallel to the direction of the earth's gravity.

4. The balancing structure of a refrigerated positioning centrifuge according to claim 3, characterized in that: Each group of the trays (11) is provided with M cavities (12), where M=12-20.

5. The balancing structure of a refrigerated positioning centrifuge according to claim 4, characterized in that: The output shaft of the motor (14) comprises an upper shaft body portion and a lower shaft body portion, a shaft sleeve is provided at the center of the bottom wall of the cross centrifugal frame (10), the cross centrifugal frame (10) is fixedly connected to the upper shaft body portion via the bottom shaft sleeve, and the lower shaft body portion abuts against the inner wall of the vibration conduction chamber (15) for transmitting vibration to the gyroscope (17) on the PCB (16) via the vibration conduction chamber (15).

6. The balancing structure of a refrigerated positioning centrifuge according to claim 5, characterized in that: The tray (11) is installed inside the centrifugal suspension (19), and the centrifugal suspension (19) is provided with hanging ears on both sides, and the hanging ears are provided with hanging holes, and the hanging holes are sleeved on the centrifugal shaft (21).

7. The balancing structure of a refrigerated positioning centrifuge according to claim 6, characterized in that: A centrifugal shaft (21) is arranged on the through hole of each group of raised wings, one end of the centrifugal shaft (21) is fixedly connected to the through hole, and the other end passes through the hanging hole, and the extension length of the other end passing through the hanging hole is less than 1 cm.

8. The balancing structure of a refrigerated positioning centrifuge according to claim 7, characterized in that: The M cavities (12) are centrally symmetrical with respect to the center of the tray (11).

9. The balancing structure of a refrigerated positioning centrifuge according to claim 8, characterized in that: It also includes a non-magnetic conductive kit (22) and a magnetic angle encoder, wherein the non-magnetic conductive kit (22) is fixedly sleeved with the lower shaft portion, and the magnetic angle encoder is arranged on the PCB board (16), and the magnetic angle encoder is located directly below the non-magnetic conductive kit (22).

10. The balancing structure of a refrigerated positioning centrifuge according to claim 9, characterized in that: The trim indication module is used to perform the following steps: S10: changing the motor speed so that after the motor is stabilized at the first speed, a sequence of rotation parameters of the gyroscope is acquired in real time, recorded as a first speed parameter sequence, wherein the rotation parameters include angular velocity, angular acceleration and angular displacement; S20: performing a fast Fourier transform on the first speed parameter sequence to obtain a corresponding frequency domain feature, recorded as a first frequency domain feature; S30: changing the motor speed so that after the motor is stabilized at the second speed, a sequence of rotation parameters of the gyroscope is acquired in real time, which is recorded as a second speed parameter sequence; S40: performing a fast Fourier transform on the second speed parameter sequence to obtain a corresponding frequency domain feature, recorded as a second frequency domain feature; S50: According to the preset first standard feature and the second standard feature, the first frequency domain feature and the second frequency domain feature are respectively calculated to obtain a first difference feature and a second difference feature; wherein the first standard feature is the frequency domain feature obtained by using the steps S10-S20 under the pre-balancing condition and is recorded as the first standard feature; the second standard feature is the frequency domain feature obtained by using the steps S30-S40 under the pre-balancing condition and is recorded as the second standard feature; S60: Inputting the obtained first difference feature and the second difference feature into a pre-trained difference calculation model to obtain a difference transformation vector; S70: performing matrix multiplication operation on the difference transformation vector and a preset basic matrix to obtain an identification matrix, wherein the identification matrix is ​​a 4×M matrix, where M is the number of serum tubes on each tray; an element of 1 in the identification matrix represents a difference serum tube, and an element of 0 represents a normal serum tube; S80: Outputting the position coordinates and weight difference estimation values ​​of the difference serum tubes in each tray according to the identification matrix.

Images