A system and method for state diagnosis and regulation of a de-dusting and sorting device based on transfer learning

Abstract

The application provides a real-time state diagnosis and regulation method and system of a threshing and separating device based on transfer learning, comprising the following steps: an information monitoring system acquires the torque of a threshing cylinder, the threshing material flow and the grain flow after screening and separation; the threshing material flow is pretreated, the pretreated threshing material flow signal is input into a cross-domain state diagnosis algorithm model based on transfer learning, and the running states of the threshing and separating device and the cleaning device are obtained; the running state of the threshing and separating device is combined with the torque of the threshing cylinder to adjust the rotation speed of the threshing cylinder and the concave plate gap; the running state of the cleaning device is combined with the grain flow under the cleaning screen to adjust the amplitude or frequency of the shaking plate; and a control system adjusts the corresponding working parts according to the rotation speed of the threshing cylinder, the concave plate gap and the amplitude or frequency of the shaking plate. The application can solve the real-time fault of a combine harvester during operation, thereby ensuring the efficient and fault-free operation of the combine harvester.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent agricultural equipment, and in particular to a threshing and separating device state diagnosis and regulation system and method based on transfer learning. BACKGROUND

[0002] In recent years, extreme weather has occurred frequently. In May 2023, Henan Province experienced a large-scale continuous rainfall, which seriously affected the maturation and harvesting of wheat. Compared with normal wheat, these wheat plants and grains have a higher water content. Due to the difficulty in threshing wet plants, the threshing and separating device often jams during the operation of the harvester in the field, affecting the harvesting progress. At the same time, with the increase in the yield of crops such as rice and wheat in China, the harvester size and weight remain unchanged, and the feeding amount is required to be as large as possible. As the "digestive system" of the combine harvester, the threshing and separating device is the most prone to failure. The increase in feeding amount, inappropriate setting of working component parameters, and high moisture content of crops all cause an increase in threshing and separating load, leading to blockage of the threshing and separating components, shortening the fault-free operation time of the combine harvester, and seriously affecting the harvesting progress. Relevant departments have repeatedly emphasized the need to improve the reliability of domestic agricultural machinery and increase the fault-free operation time of the harvester in the field. Therefore, there is an urgent need for a method that can effectively monitor and control the threshing and separating device of the combine harvester in real time, allowing the combine harvester to work efficiently, reducing the occurrence of faults, improving the reliability of domestic agricultural machinery, and promoting the intelligent and high-end development of agricultural machinery.

[0003] Currently, most research on monitoring the threshing and separating device uses torque sensors, vibration sensors, etc. to directly monitor the threshing cylinder, and the monitoring method is relatively single. In addition, the harvester body vibrates greatly during field operation, which affects the monitoring effect. The research on the working state diagnosis algorithm of the threshing and separating device often uses machine learning methods, and it is not possible to artificially create faults to collect fault original signals during the operation of the harvester in the field. This is time-consuming and labor-intensive, and will affect the harvesting progress, so the data samples for training the model are mostly from bench tests. Bench tests can only simulate the change in the operating state of the threshing and separating components caused by a single or a small number of reasons, and cannot accurately simulate the state of the harvester in the field. Therefore, the model obtained from the fault signals obtained from bench tests often has poor effect when applied to actual field operation.

[0004] The invention patent with publication number CN111310830B discloses a combine harvester blockage fault diagnosis system and method. The system includes a sensor module, a data acquisition module, an interactive display screen, an audible and visual alarm device, and an embedded industrial computer. The sensor module is used to monitor various parameter information of the harvester online. The data monitored by the sensor module is collected by the data acquisition module and then processed by the embedded industrial computer. First, the SDAE-BP model is trained using the collected data to obtain the weight parameters and bias parameters of the SDAE-BP model. Then, the SDAE-BP model is pre-prepared in the embedded industrial computer. After that, the data collected can be used to determine the running state of the combine harvester by the SDAE-BP model in the embedded industrial computer. However, only acceleration sensors are used for monitoring, which is a relatively single method, and the useful signals that can be obtained are limited, and the method is easily affected by the vibration noise of the machine body.

[0005] The invention patent with publication number CN113945380A discloses a threshing cylinder bearing vibration fault analysis method and system and a combine harvester. When the data acquisition system detects the threshing cylinder speed fluctuation, the vibration signal is transmitted to the data processing system. The data processing system processes the received vibration signal and extracts fault information. The fault signal is analyzed and judged. According to the signal characteristics, the position of the impact or fault is determined to be in the threshing cylinder or in the remaining components other than the threshing cylinder. The fault signal data is stored in the memory. At the same time, the fault warning platform displays and alarms according to the results of the signal processing system. For the bearing running state detection of the threshing cylinder, the data acquisition is convenient. The known deterministic component signals and random impact signals in the vibration signal are decomposed to eliminate the interference of irrelevant factors on fault monitoring. The signal is effectively analyzed and processed to accurately locate the fault position. However, the vibration signal of the harvester itself overlaps with the vibration signal of the threshing cylinder, and the vibration signal of the harvester itself changes with the power of the engine. Therefore, it is difficult to accurately separate the vibration signal of the threshing cylinder from the original signal, and the diagnosis accuracy is low.

[0006] The invention patent with publication number CN113033833A discloses a combine harvester threshing cylinder harvesting state fault diagnosis method, which comprises the following steps: sample acquisition and pretreatment, training of a network pre-model for source domain data samples, supervised fine-tuning of a prediction model for a small number of labeled target domain samples, and output of a diagnosis result. The prediction model established can update the weights and bias values layer by layer to express the input signal in stages, which can more effectively make correct diagnosis for cylinder faults. The migration of weights and bias values from the source domain to the target domain is used to adapt to new target sample recognition, ultimately achieving the effect of improving the target domain sample fault recognition accuracy. By combining the advantages of the transfer learning method in solving different domain samples, the problem of not being able to accurately identify sample faults when the working state samples of the threshing cylinder are insufficient due to many factors is overcome, and the target domain sample fault recognition accuracy is improved. However, this method can only perform well when the source domain and target domain data come from the same feature distribution space, i.e., subject to the same distribution assumption. However, during field operation of the harvester, the working conditions of the threshing cylinder are diverse, and the source domain and target domain data generally follow different distributions, making it very difficult to collect a large amount of labeled data with fault information and subject to the same distribution assumption.

[0007] The invention patent with publication number CN113033833A discloses a combine harvester threshing cylinder fault simulation monitoring system and method. The monitoring system comprises a brake loading device, a fault auxiliary disc and a data monitoring assembly. The fault auxiliary disc is mounted on the threshing cylinder, and the output end of the brake loading device is connected to the shaft end of the threshing cylinder. The data monitoring assembly is arranged at the connection between the brake loading device and the shaft end of the threshing cylinder, and the brake loading device is electrically connected to the data monitoring assembly. By pre-installing the fault auxiliary disc and the brake loading device on the threshing cylinder, and using the data monitoring assembly to collect threshing related parameters, the brake loading device can control the dynamic load of the threshing cylinder. The dynamic imbalance fault of the threshing cylinder, the shaft end bearing fault and the cylinder blockage fault can be pre-set on the whole machine. However, there are many reasons for the threshing cylinder fault during field operation of the harvester. Relying solely on the pre-installed fault auxiliary disc and brake loading device cannot completely simulate the actual fault condition, and the data monitoring assembly only monitors the connection between the brake loading device and the shaft end of the threshing cylinder, and the information obtained is insufficient. SUMMARY

[0008] In view of the deficiencies in the prior art, the present application provides a kind of based on the state diagnosis and regulation system and method of migration learning of threshing and separating device, the flow variation of threshing product is monitored, so as to reflect the running state of threshing and separating device;For the working state diagnosis method of combined harvester threshing and separating device, most of the fault data is obtained by bench test, the diagnostic model obtained has poor effect in application to actual scene working state diagnosis and working component parameter regulation problem, a kind of based on the state diagnosis and regulation method of migration learning of threshing and separating device is proposed to solve the real-time fault when combined harvester is working, so as to ensure the efficient and fault-free operation of combined harvester.

[0009] The present application achieves the above technical object by the following technical means.

[0010] A kind of based on the real-time state diagnosis and regulation method of migration learning of threshing and separating device, including the following steps:

[0011] Step S1: when combined harvester is field working, information monitoring system real-time obtains the torque of threshing cylinder, threshing product flow and grain flow after screening, and transmits monitoring information to embedded processor;

[0012] Step S2: the threshing product flow is preprocessed, and the preprocessed threshing product flow signal is input into the cross-domain state diagnosis algorithm model based on migration learning, to obtain the running state of threshing and separating device;

[0013] Step S3: the running state of threshing and separating device is combined with the torque of threshing cylinder, and the speed of threshing cylinder and concave clearance are adjusted;The running state of cleaning device is combined with the grain flow under the cleaning screen, and the amplitude or frequency of shaking plate is adjusted;

[0014] Step S4: control system adjusts corresponding working components according to the speed of threshing cylinder, concave clearance and the amplitude or frequency of shaking plate.

[0015] Further, the cross-domain state diagnosis algorithm model based on migration learning described in step S2 is established, specifically including the following steps:

[0016] Step S2.1: obtain the data set required for model training, specifically:

[0017] Step S2.1.1: establish threshing and separating device bench, install shaking plate on the bench, and pretreat threshing product flow signal collected by test bench as source domain data set;

[0018] Step S2.1.2: install shaking plate on combined harvester for field operation, and pretreat threshing product flow signal collected during field operation as target domain data set;

[0019] Step S2.1.3: Labeling the source domain dataset and the target domain dataset, the operating states of the threshing and separating device and the cleaning device include normal operating states and abnormal operating states;

[0020] Step S2.1.4: Using the method of overlap sampling to perform data enhancement on the source domain and the target domain dataset, increasing the number of training dataset;

[0021] Step S2.2: Building a model, training using the dataset, obtaining a real-time state diagnosis model of the threshing and separating device based on transfer learning, specifically:

[0022] Step S2.2.1: Establishing a multi-layer adaptive convolutional neural network model;

[0023] Step S2.2.2: Inputting the labeled source domain dataset and the unlabeled target domain dataset into the multi-layer adaptive convolutional neural network model for training, ending the training when the loss tends to be stable, and obtaining the real-time state diagnosis model of the threshing and separating device based on transfer learning;

[0024] Step S2.3: Testing the real-time state diagnosis model of the threshing and separating device based on transfer learning, verifying its diagnosis accuracy, specifically as follows:

[0025] Step S2.3.1: Inputting the target domain dataset into the real-time state diagnosis model of the threshing and separating device based on transfer learning, obtaining the diagnosis result;

[0026] Step S2.3.2: Comparing the diagnosis result with the true label, and calculating the diagnosis accuracy.

[0027] Further, a multi-layer adaptive convolutional neural network model is established, specifically including the following steps:

[0028] Step S2.2.1.1: Using a wide convolution kernel to preliminarily extract shallow features of the threshing and separating device;

[0029] Step S2.2.1.2: Using a double-scale convolution module with a residual structure to extract features of the threshing and separating device at multiple scales, the double scale being a 3x1 convolution kernel and a 5x1 convolution kernel;

[0030] Step S2.2.1.3: Using a continuous 3x1 convolution kernel to extract deep features of the threshing and separating device;

[0031] Step S2.2.1.4: Using adaptive batch normalization to replace the mean μ s and the variance σ s of all source domain datasets in the model with the mean μ t and the variance σ t of the target domain dataset;

[0032] Step S2.2.1.5: Calculate the maximum mean difference between the source domain and the target domain dataset using the multi-kernel maximum average distance, and then combine it with the classification loss, the formula of the multi-kernel maximum average distance is:

[0033]

[0034] Wherein: X s is the source domain dataset; X t is the target domain dataset; X i s is the i-th source domain data in the source domain dataset; X i t is the i-th target domain data in the target domain dataset; n s is the number of source domain datasets; n t is the number of target domain datasets; H is the reproducing kernel Hilbert space; k is the Gaussian kernel function;

[0035] The loss function is calculated as follows:

[0036]

[0037]

[0038] Wherein, Loss(θ) is the loss function of the multi-layer adaptive convolutional neural network, is the cross-entropy loss of the source domain sample, is the training sample, is the true label of the target domain sample, is the diagnosis result, θ is the parameter set, and λ≥0 is the weight parameter;

[0039] Step S2.2.1.6: Classify the extracted features through the full connection layer to realize the running state diagnosis of the threshing and separating device and the cleaning device.

[0040] Further, the discharge flow is pretreated, specifically:

[0041] The surface array of the shaking plate is installed with m*n sensors, and m*n discharge flow signals are collected, which are respectively: w 11 , w 12 ……w 1n ,……w m1 ,……w mn ;

[0042] Taking 1s as the time sequence segment length, the segments overlap by 50%, each discharge flow signal is sliced, and then a (m*n) *1 matrix is formed, so that 2n-1 matrix discharge flow signals can be obtained from n second discharge flow signals;

[0043] The 2n-1 matrix discharge flow signals are filtered to eliminate the influence of vibration noise.

[0044] Further, the threshing cylinder rotating speed and the concave clearance are adjusted in combination with the operating state of the threshing and separating device and the torque of the threshing cylinder, specifically:

[0045] When the operating state of the threshing and separating device is normal, if the torque of the threshing cylinder is less than the normal value, the feeding amount is small at this time, and the threshing cylinder is in a low load state, so the threshing cylinder rotating speed is reduced and the concave clearance is decreased.

[0046] When the operating state of the threshing and separating device is normal, if the torque of the threshing cylinder is greater than the normal value, the feeding amount is large at this time, and the threshing cylinder is in a high load state, so the threshing cylinder rotating speed is increased and the concave clearance is increased.

[0047] When the operating state of the threshing and separating device is abnormal, if the torque of the threshing cylinder is greater than the normal value and is in a continuously increasing state, it is in a clogging tendency state at this time, so the threshing cylinder rotating speed is increased and the concave clearance is increased.

[0048] When the operating state of the threshing and separating device is abnormal, if the torque of the threshing cylinder is greater than the normal value and is stable, it is in a clogging state at this time, so an alarm is given and the operation is stopped.

[0049] Further, the frequency or amplitude of the shaking plate is adjusted in combination with the operating state of the cleaning device and the grain flow under the cleaning screen, specifically:

[0050] When the operating state of the cleaning device is normal, the frequency or amplitude of the shaking plate remains unchanged.

[0051] When the operating state of the cleaning device is abnormal, if the grain flow under the cleaning screen is less than or equal to the normal range, the frequency or amplitude of the shaking plate is increased.

[0052] When the operating state of the cleaning device is abnormal, if the grain flow under the cleaning screen decreases rapidly from the normal range, it is in a clogging tendency state at this time, so the frequency or amplitude of the shaking plate is increased, the frequency or amplitude of the fish scale screen is increased, and the opening degree of the fish scale screen is increased.

[0053] When the operating state of the cleaning device is abnormal, if the grain flow under the cleaning screen is zero, it is in a clogging state at this time, so an alarm is given and the operation is stopped.

[0054] A system of a real-time state diagnosis and regulation method of a threshing and separating device based on transfer learning, comprising a threshing and separating device, a cleaning device, an information monitoring system, an embedded processor and a control system.

[0055] The threshing and separating device is provided with a cleaning device at the bottom; the threshing and separating device comprises a concave screen with adjustable gap; the cleaning device comprises a shaking plate, a fish scale screen, a woven screen, a grain flow sensor and a vibration generator; the shaking plate and the fish scale screen are respectively arranged at both sides of the outlet of the threshing and separating device, and the shaking plate is arranged above the fish scale screen; the shaking plate and the fish scale screen are respectively connected with the vibration generator; the fish scale screen is provided with the woven screen at the bottom; the shaking plate is provided with the sensor inside, which is used for monitoring the threshing material flow falling into the shaking plate; the grain flow sensor is arranged at the bottom of the woven screen, which is used for measuring the grain flow under the cleaning screen;

[0056] The information monitoring system comprises at least one sensor, which is used for monitoring the torque and rotating speed of the threshing cylinder; the information monitoring system obtains the grain flow under the cleaning screen and the threshing material flow; the information monitoring system inputs the collected and monitored information into the embedded processor, which is used for diagnosing the state of the threshing and separating device and the cleaning device; the control system adjusts the rotating speed of the threshing cylinder, the gap of the concave screen and the amplitude or frequency of the shaking plate according to the state of the threshing and separating device and the cleaning device.

[0057] Further, the shaking plate comprises an array shaking plate, a second shaking plate support, a sliding block, a base, a first damping and a spoke; one end of the array shaking plate is provided with the spoke, which extends above the fish scale screen; the array shaking plate is arranged on the second shaking plate support through the first damping; the second shaking plate support is arranged on the sliding block, which is movably arranged in the sliding rail with slope of the base; the sliding block is reciprocated in the sliding rail with slope through the actuator, so that the second shaking plate support is reciprocated in height, which is used for changing the amplitude or frequency of the array shaking plate; the surface of the array shaking plate is provided with the sensor, which is used for monitoring the threshing material flow falling into the array shaking plate.

[0058] Further, the array shaking plate comprises a shaking top plate, a piezoelectric sensor, a second damping and a first shaking plate support; a plurality of shaking top plates are arrayed above the first shaking plate support; each shaking top plate is connected with the first shaking plate support through the second damping; the piezoelectric sensor is arranged at the center of the bottom surface of each shaking top plate, which is used for monitoring the threshing material flow falling into the shaking top plate.

[0059] Further, the control system comprises a threshing cylinder rotating speed controller, a concave gap controller and a shaking plate moving controller; the threshing cylinder rotating speed controller adjusts the rotating speed of the threshing cylinder according to the state of the threshing and separating device and the cleaning device; the concave gap controller adjusts the gap of the concave screen according to the state of the threshing and separating device and the cleaning device; the shaking plate moving controller adjusts the moving speed and reciprocating frequency of the sliding block according to the state of the threshing and separating device and the cleaning device.

[0060] The present application has the following advantages:

[0061] 1. The real-time state diagnosis and regulation method of the threshing and separating cleaning device based on transfer learning, which monitors the changes of the threshed material through the shaking plate, so that the working state of the threshing cylinder is reflected through the change characteristics of the threshed material, and the cleaning load state of the cleaning device can also be obtained from the material on the shaking plate.

[0062] 2. The real-time state diagnosis and regulation system of the threshing and separating cleaning device based on transfer learning, which improves the monitoring accuracy of the threshed material through array design, and reduces the influence of machine body vibration on monitoring through multiple layers of damping. At the same time, the shaking plate can reciprocate in the slide rail, and when the material on the shaking plate is accumulated, the shaking plate can be accelerated to improve the cleaning efficiency and prevent the cleaning screen from being blocked.

[0063] 3. The real-time state diagnosis and regulation method of the threshing and separating cleaning device based on transfer learning, which uses multiple sensors for monitoring, realizes multi-source sensor information fusion, can prevent errors caused by single sensor monitoring, leads to incorrect diagnosis of the running state of the threshing and separating cleaning device by the diagnosis algorithm, and further affects the working condition of the whole machine.

[0064] 4. The real-time state diagnosis and regulation method of the threshing and separating cleaning device based on transfer learning, most of the abnormal working signals of the threshing and separating cleaning device are collected in the bench test, and the model trained based on the transfer learning of the cross-domain state diagnosis algorithm is used to achieve good results in the actual operation of the harvester in the field.

[0065] 5. The real-time state diagnosis and regulation method of the threshing and separating cleaning device based on transfer learning, the cross-domain state diagnosis algorithm based on transfer learning uses the batch normalization method to reduce the domain shift of the network in the shallow layer, and uses the multi-core maximum average distance to reduce the domain shift of the network in the deep layer, so that the model shows good diagnosis effect on the target domain.

[0066] 6. The real-time state diagnosis and regulation method of the threshing and separating cleaning device based on transfer learning, the real-time state diagnosis method based on transfer learning can diagnose the real-time fault of the threshing and separating device and the cleaning device of the combine harvester and give the regulation strategy, so that the combine harvester can work efficiently and faultlessly, and is suitable for current manual driving operation and future unmanned operation scenes. BRIEF DESCRIPTION OF DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. The drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0068] Figure 1 The schematic diagram of the real-time state diagnosis and regulation system of the threshing and separating device based on the transfer learning.

[0069] Figure 2 The block diagram of the real-time state diagnosis and regulation system of the threshing and separating device based on the transfer learning.

[0070] Figure 3 The shaft drawing of the shaking plate.

[0071] Figure 4 The 2x2 array type distributed array shaking plate in the embodiments of the present application.

[0072] Figure 5 The second shaking plate bracket 2 installation position drawing.

[0073] Figure 6 The shaft drawing of the array shaking plate.

[0074] Figure 7 The shaft drawing of the slider and the base.

[0075] Figure 8 The schematic diagram of the slider reciprocating in the base.

[0076] Figure 9 The schematic diagram of the information monitoring system.

[0077] Figure 10 The schematic diagram of the control system.

[0078] Figure 11 The flow chart of the real-time state diagnosis and regulation method of the threshing and separating device based on the transfer learning.

[0079] Figure 12 The working flow chart of the multi-layer adaptive convolutional neural network.

[0080] Figure 13 The logic diagram of the determination of the threshing cylinder rotating speed and the recess plate gap adjustment.

[0081] Figure 14 The logic diagram of the shaking plate adjustment.

[0082] In the drawings:

[0083] 1 - threshing and separating device; 1 -1 - concave screen; 2 - cleaning device; 2-1 - shaking plate; 2-1 -1 - array shaking plate; 2-1 -1 -1 - shaking top plate; 2-1 -1 -2 - piezoelectric sensor; 2-1 -1 -3 - first shaking plate support; 2-1 -1 -4 - second shock absorbing damping; 2-1 -2 - second shaking plate support; 2-1 -3 - sliding block; 2-1 -4 - base; 2-1 -5 - shock absorbing damping; 2-1 -6 - spoke; 2-2 - fish scale screen; 2-3 - woven screen; 2-4 - grain flow sensor; 3 - information monitoring system; 3-1 - torque sensor; 4 - embedded processor; 5 - control system. DETAILED DESCRIPTION

[0084] Embodiments of the present application are described in detail below with reference to the attached drawings, wherein like or similar elements are denoted by the same or similar reference signs throughout the drawings. The embodiments described below are examples for explaining the present application and are not intended to be limiting of the present application.

[0085] In the description of the present application, it is to be understood that the orientations or positional relationships indicated by the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "axial", "radial", "vertical", "horizontal", "inner", "outer", etc. are based on the orientations or positional relationships shown in the drawings, and are merely intended to facilitate the description of the present application and simplify the description, and are not intended to indicate or imply that the devices or elements indicated by the orientations must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise explicitly specified and limited.

[0086] In the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connecting", "fixing" and the like should be understood broadly, for example, can be fixedly connected, can be detachably connected, or integrally connected; can be mechanically connected, or electrically connected; can be directly connected, or indirectly connected through an intermediate medium, or can be the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0087] As Figure 1 andFigure 2 As shown, the system for real-time state diagnosis and regulation of the threshing and separating device based on transfer learning of the application comprises a threshing and separating device 1, a cleaning device 2, an information monitoring system 3, an embedded processor 4 and a control system 5; the threshing and separating device 1 is installed at the bottom of the cleaning device 2; the threshing and separating device 1 comprises an adjustable gap concave screen 1-1; the adjustable gap concave screen 1-1 is a prior art in the threshing and separating device 1, and thus its structure is not described. The cleaning device 2 comprises a shaking plate 2-1, a fish scale screen 2-2, a woven screen 2-3, a grain flow sensor 2-4 and a vibration generator; the threshing and separating device 1 is installed at the outlet of the cleaning device 2, and the shaking plate 2-1 and the fish scale screen 2-2 are respectively installed on both sides of the threshing and separating device 1, and the shaking plate 2-1 is located above the fish scale screen 2-2; the shaking plate 2-1 and the fish scale screen 2-2 are respectively connected with the vibration generator; the vibration generator in the combine harvester is generally an eccentric wheel, and the fixed frequency and amplitude vibration is realized by the rotation of the eccentric wheel; the fish scale screen 2-2 is provided with the woven screen 2-3 at the bottom; the shaking plate 2-1 is provided with a sensor for monitoring the flow of the threshed material falling into the shaking plate 2-1; the grain flow sensor 2-4 is located at the bottom of the woven screen 2-3 for measuring the grain flow under the cleaning screen;

[0088] The information monitoring system 3 comprises at least one sensor for monitoring the threshing cylinder torque and rotating speed; the information monitoring system 3 obtains the grain flow under the cleaning screen and the threshed material flow; the information monitoring system 3 inputs the collected and monitored information into the embedded processor 4 for real-time diagnosis of the state of the threshing and separating device 1 and the cleaning device 2; the control system 5 adjusts the rotating speed of the threshing cylinder, the concave gap and the amplitude or frequency of the shaking plate 2-1 according to the corresponding strategy of the state of the threshing and separating device 1 and the cleaning device 2, so that the combine harvester works efficiently and without failure.

[0089] As shown Figure 3 , Figure 6 , Figure 7 and Figure 8As shown, the shaking plate 2-1 includes an array shaking plate 2-1-1, a second shaking plate support 2-1-2, a sliding block 2-1-3, a base 2-1-4, a first damping 2-1-5 and a spoke 2-1-6; one end of the array shaking plate 2-1-1 is installed with the spoke 2-1-6, the spoke 2-1-6 extends above the fish scale screen 2-2 to reduce the influence of the vibration of the second shaking plate support 2-1-2 on the array shaking plate 2-1-1; the array shaking plate 2-1-1 is installed on the second shaking plate support 2-1-2 through the first damping 2-1-5; the second shaking plate support 2-1-2 is installed on the sliding block 2-1-3, the sliding block 2-1-3 is movably installed in the sliding rail with slope of the base 2-1-4, the sliding block 2-1-3 reciprocates in the sliding rail with slope through the actuator, the second shaking plate support 2-1-2 reciprocates in height, and the amplitude or frequency of the array shaking plate 2-1-1 is changed; the base 2-1-4 is installed in the cleaning device 2. The surface of the array shaking plate 2-1-1 is provided with a sensor for monitoring the flow of the detached objects falling on the array shaking plate 2-1-1. The threshed and separated grains of the threshing and separating device 1 fall on the shaking plate 2-1, the shaking plate 2-1 can uniformly screen the grains and monitor the change of the detached objects above the shaking plate at the same time; when the control system 5 does not control the actuator, the sliding block 2-1-3 is fixed in the sliding rail, at this time, the shaking plate 2-1 moves with the same frequency and amplitude as the fish scale screen 2-2; when the actuator controls the sliding block 2-1-3 to reciprocate in the sliding rail according to the set speed and amplitude, the array shaking plate 2-1-1 reciprocates in height, at this time, the frequency and amplitude of the shaking plate 2-1 increase, thereby accelerating the movement of the grains on the shaking plate to the fish scale screen 2-2, slowing down the accumulation of the materials on the shaking plate 2-1, improving the screening efficiency and accelerating the screening speed. The general actuator is a cylinder, then the amplitude of the sliding block 2-1-3 in the sliding rail can be understood as the moving distance of the sliding block 2-1-3, the greater the moving distance of the sliding block 2-1-3, the greater the change in height of the array shaking plate 2-1-1 driven by the sliding block 2-1-3.

[0090] The array shaking plate 2-1-1 includes a shaking top plate 2-1-1-1, a piezoelectric sensor 2-1-1-2, a second damping 2-1-1-4 and a first shaking plate support 2-1-1-3; a plurality of shaking top plates 2-1-1-1 are arrayed above the first shaking plate support 2-1-1-3, each shaking top plate 2-1-1-1 is connected with the first shaking plate support 2-1-1-3 through the second damping 2-1-1-4; the piezoelectric sensor 2-1-1-2 is installed at the center of the bottom surface of each shaking top plate 2-1-1-1 for monitoring the flow of the detached objects falling on the shaking top plate 2-1-1-1. The array shaking plate 2-1-1 is arranged in an m*n array, wherein m can be equal to n, such as Figure 4As shown in the figure, the 2×2 array distribution is taken as an example, each array shaking plate 2-1-1 is not closely connected, but has a 1mm spacing, preventing the vibration of itself from affecting other array shaking plates 2-1-1, thereby causing errors in the monitoring of the discharges; the total number of spokes 2-1-6 is j, and j / n spokes 2-1-6 are arranged on each array shaking plate 2-1-1; the embedded processor can obtain the spatial distribution information of the discharges by processing the monitoring information of the array shaking plates arranged in an array, and can draw a distribution diagram of the discharges falling on the shaking plate.

[0091] As shown in the figure, Figure 5 The left figure is a 2×2 array distribution array shaking plate 2-1-1, and the right figure is a 4×4 array distribution array shaking plate 2-1-1; the array shaking plate 2-1-1 is installed on the second shaking plate support 2-1-2 through the shock damping 2-1-5; the number of the second shock damping 2-1-1-4 is m×n; the cross beam of the second shaking plate support 2-1-2 is n, and the vertical beam is m-1, thereby increasing the stability of the support.

[0092] As shown in the figure, Figure 7 The base 2-1-4 is provided with a sliding groove, and the sliding block 2-1-3 can move in the sliding groove; the upper part of the sliding groove is not completely open, and is provided with a limiting strip to prevent the sliding block 2-1-3 from falling out of the sliding groove; the lower part of the sliding groove has a completely open opening, so that when a grain accidentally falls into the sliding groove, it can slide out of the opening, avoiding the accumulation of grains in the sliding groove and causing blockage.

[0093] As shown in the figure, Figure 9 The information monitoring system 3 includes a torque sensor 3-1, which monitors the threshing cylinder load; the information monitoring system 3 obtains the discharge flow information monitored by the shaking plate 2-1; the information monitoring system 3 obtains the screen flow information monitored by the screen flow sensor 2-4; the information monitored by the information monitoring system is displayed on the display device, providing a reference for the driver.

[0094] As shown in the figure, Figure 10 The control system 5 includes a threshing cylinder speed controller, a concave clearance controller and a shaking plate moving controller; the threshing cylinder speed controller adjusts the threshing cylinder speed according to the state of the threshing and separating device 1 and the cleaning device 2; the concave clearance controller adjusts the concave clearance according to the state of the threshing and separating device 1 and the cleaning device 2; the shaking plate moving controller adjusts the moving speed and reciprocating frequency of the sliding block according to the state of the threshing and separating device 1 and the cleaning device 2.

[0095] As shown in the figure, Figure 11 The real-time state diagnosis and regulation method of the threshing and separating device based on transfer learning, comprising the following steps:

[0096] Step S1: When the combine harvester is working in the field, the information monitoring system 3 obtains the torque of the threshing cylinder, the threshing flow and the cleaned grain flow in real time, and transmits the monitoring information to the embedded processor 4;

[0097] Step S2: The threshing flow is preprocessed, and the preprocessed threshing flow signal is input into the cross-domain state diagnosis algorithm model based on transfer learning to obtain the running state of the threshing and separating device 1 and the cleaning device 2;

[0098] The threshing flow is preprocessed, specifically:

[0099] The surface of the shaking plate 2-1 is arrayed with m×n sensors, and m×n threshing flow signals are collected, which are respectively: 11 , w 12 ……w 1n ,……w m1 ,……w mn ;

[0100] With 1s as the time sequence segment length, the segments overlap by 50%, each threshing flow signal is sliced, and then a (m×n)×1 matrix is formed, so that 2n-1 matrix threshing flow signals can be obtained from n second threshing flow signals.

[0101] The 2n-1 matrix threshing flow signals are filtered to eliminate the influence of vibration noise.

[0102] The cross-domain state diagnosis algorithm model based on transfer learning in step S2 is established, which specifically includes the following steps:

[0103] Step S2.1: Obtain the data set required for model training, specifically:

[0104] Step S2.1.1: Establish a threshing and cleaning device test bench, install the shaking plate 2-1 on the test bench, and preprocess the threshing flow signal collected by the test bench as the source domain data set;

[0105] Step S2.1.2: Install the shaking plate 2-1 on the combine harvester for field work, and preprocess the threshing flow signal collected during field work as the target domain data set;

[0106] Step S2.1.3: Label the source domain data set and the target domain data set, and the running state of the threshing and separating device 1 and the cleaning device 2 includes normal running state and abnormal running state;

[0107] Step S2.1.4: Use the overlapping sampling method to perform data enhancement on the source domain and target domain data sets to increase the number of training data sets;

[0108] Step S2.2: building a model, training using the data set, obtaining a real-time state diagnosis model of the dehulling and cleaning device based on transfer learning, specifically:

[0109] Step S2.2.1: establishing a multi-layer adaptive convolutional neural network model, as shown in FIG. 2, specifically including the following steps: Figure 12

[0110] Step S2.2.1.1: using a wide convolution kernel to preliminarily extract shallow features of the dehulling and cleaning device flow signal;

[0111] Step S2.2.1.2: using a double-scale convolution module with a residual structure to extract features of the dehulling and cleaning device flow signal at multiple scales, the double scale being a 3x1 convolution kernel and a 5x1 convolution kernel;

[0112] Step S2.2.1.3: using a continuous 3x1 convolution kernel to extract deep features of the dehulling and cleaning device flow signal;

[0113] Step S2.2.1.4: using adaptive batch normalization to replace the mean μ s and variance σ s of all source domain data sets in the model with the mean μ t and variance σ t of the target domain data set;

[0114] Step S2.2.1.5: using a multi-core maximum average distance to calculate the maximum mean difference between the source domain and the target domain data set, and then combining it with the classification loss, the formula of the multi-core maximum average distance being:

[0115]

[0116] wherein X s is the source domain data set; X t is the target domain data set; X i s is the i th source domain data in the source domain data set; X i t is the i th target domain data in the target domain data set; n s is the number of source domain data sets; n t is the number of target domain data sets; H is a reproducing kernel Hilbert space; and k is a Gaussian kernel function.

[0117] The loss function is calculated as follows:

[0118]

[0119] ​

[0120] Where Loss(θ) is the loss function of the multilayer adaptive convolutional neural network. The cross-entropy loss is the value of the source domain samples. As training samples, The true labels for the target domain samples. For the diagnostic results, θ is the parameter set, and λ≥0 is the weight parameter;

[0121] Step S2.2.1.6: Classify the extracted features through a fully connected layer to achieve operational status diagnosis of the threshing and separation device 1 and the cleaning device 2.

[0122] Step S2.2.2: Input the labeled source domain dataset and the unlabeled target domain dataset into the multilayer adaptive convolutional neural network model for training. When the loss tends to stabilize, the training ends, and the real-time status diagnosis model of the sorting and cleaning device based on transfer learning is obtained.

[0123] Step S2.3: Test the real-time status diagnosis model of the sorting and cleaning device based on transfer learning to verify its diagnostic accuracy, as follows:

[0124] Step S2.3.1: Input the target domain dataset into the real-time status diagnosis model of the sorting and cleaning device based on transfer learning to obtain the diagnosis results;

[0125] Step S2.3.2: Compare the diagnostic results with the true labels and calculate the diagnostic accuracy.

[0126] Step S3: Adjust the threshing drum speed and concave plate gap based on the operating status of the threshing and separating device 1 and the torque of the threshing drum, such as... Figure 13 As shown, specifically:

[0127] When the threshing and separating device 1 is operating normally, if the torque of the threshing drum is less than the normal value, the feed rate is small and the threshing drum is under low load. It is necessary to reduce the speed of the threshing drum and reduce the gap between the concave plates.

[0128] When the threshing and separating device 1 is operating normally, if the torque of the threshing drum is greater than the normal value, the feed rate is large and the threshing drum is under high load. It is necessary to increase the speed of the threshing drum and increase the gap between the concave plates.

[0129] When the threshing and separating device 1 is in an abnormal operating state, if it is in an unloaded state at this time, the operation of the threshing drum should be stopped if harvesting is not to continue.

[0130] When the threshing and separating device 1 is in an abnormal operating state, if the torque of the threshing drum is greater than the normal value and continues to increase, then it is in a state of clogging and needs to increase the speed of the threshing drum and increase the gap between the concave plates.

[0131] When the running state of the threshing separation device 1 is abnormal, if the torque of the threshing cylinder is greater than the normal value and is stable, it is a clogging state at this time, which needs to be alarmed and stopped for manual inspection.

[0132] The amplitude or frequency of the shaking plate 2-1 under the cleaning device 2 is adjusted in combination with the running state of the cleaning device 2 and the grain flow under the cleaning screen, as shown in Figure 14 , specifically:

[0133] When the running state of the cleaning device 2 is normal, the frequency or amplitude of the shaking plate 2-1 remains unchanged;

[0134] When the running state of the cleaning device 2 is abnormal, if the grain flow under the cleaning screen is less than or equal to the normal range, the frequency or amplitude of the shaking plate 2-1 is increased;

[0135] When the running state of the cleaning device 2 is abnormal, if the grain flow under the cleaning screen decreases from the normal range, it is a clogging state at this time, the frequency or amplitude of the shaking plate 2-1 is increased, the frequency or amplitude of the fish scale screen 2-2 is increased, and the opening of the fish scale screen 2-2 is increased;

[0136] When the running state of the cleaning device 2 is abnormal, if the grain flow under the cleaning screen is zero, it is a clogging state at this time, which needs to be alarmed and stopped.

[0137] Step S4: The control system 5 adjusts the corresponding working parts according to the threshing cylinder speed, the concave clearance and the amplitude or frequency of the shaking plate 2-1.

[0138] It should be understood that although the present specification is described according to each embodiment, not every embodiment contains only one independent technical solution, and the description manner of the specification is only for clarity, and the skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be combined to form other embodiments that can be understood by the skilled in the art.

[0139] The series of detailed descriptions listed above are only specific descriptions of the feasible embodiments of the present application, and are not used to limit the protection scope of the present application, and any equivalent embodiments or changes made without departing from the spirit of the present application should be included in the protection scope of the present application.

Claims

1. A method for real-time state diagnosis and regulation of a de-entanglement and separation device based on transfer learning, characterized in that, Comprising the following steps: Step S1: When the combine harvester is working in the field, the information monitoring system (3) obtains the torque of the threshing cylinder, the threshed material flow and the cleaned grain flow after screening in real time, and transmits the monitoring information to the embedded processor (4); Step S2: The threshed material flow is preprocessed, and the preprocessed threshed material flow signal is input into the cross-domain state diagnosis algorithm model based on transfer learning to obtain the running state of the threshing and separating device (1) and the cleaning device (2); Step S3: The rotation speed of the threshing cylinder and the concave clearance are adjusted in combination with the running state of the threshing and separating device (1) and the torque of the threshing cylinder; the amplitude or frequency of the shaking plate (2-1) is adjusted in combination with the running state of the cleaning device (2) and the grain flow under the cleaning screen; Step S4: The control system (5) adjusts the corresponding working components according to the rotation speed of the threshing cylinder, the concave clearance and the amplitude or frequency of the shaking plate (2-1); The cross-domain state diagnosis algorithm model based on transfer learning described in step S2 is established, specifically comprising the following steps: Step S2.1: Obtain the data set required for model training, specifically: Step S2.1.1: Establish a threshing and cleaning device bench, install the shaking plate (2-1) on the bench, and preprocess the threshed material flow signal collected by the test bench as the source domain data set; Step S2.1.2: Install the shaking plate (2-1) on the combine harvester for field work, and preprocess the threshed material flow signal collected during field work as the target domain data set; Step S2.1.3: Label the source domain data set and the target domain data set, the running state of the threshing and separating device (1) and the cleaning device (2) includes normal running state and abnormal running state; Step S2.1.4: Use the overlapping sampling method to perform data enhancement on the source domain and target domain data sets to increase the number of training data sets; Step S2.2: Build a model and use the data set to train to obtain a real-time state diagnosis model of the threshing and separating device based on transfer learning, specifically: Step S2.2.1: Establish a multi-layer adaptive convolutional neural network model; Step S2.2.2: Input the labeled source domain data set and the unlabeled target domain data set into the multi-layer adaptive convolutional neural network model for training, and end the training when the loss tends to be stable to obtain the real-time state diagnosis model of the threshing and separating device based on transfer learning; Step S2.3: Test the real-time state diagnosis model of the threshing and separating device based on transfer learning to verify its diagnosis accuracy, specifically as follows: Step S2.3.1: Input the target domain data set into the real-time state diagnosis model of the threshing and separating device based on transfer learning to obtain the diagnosis result; Step S2.3.2: Compare the diagnosis result with the true label to calculate the diagnosis accuracy.

2. The real-time state diagnosis and regulation method of a migration learning-based de- cleaning device according to claim 1, characterized in that, The multi-layer adaptive convolutional neural network model is established, specifically comprising the following steps: Step S2.2.1.1: Use a wide convolution kernel to preliminarily extract the shallow features of the threshed material flow signal; Step S2.2.1.2: Extracting features of the extraneous matter flow signal at multiple scales using a dual-scale convolution module with residual structure, the dual-scale being a 3x1 convolution kernel and a 5x1 convolution kernel; Step S2.2.1.3: Extracting deep-level features of the extraneous matter flow signal using a continuous 3x1 convolution kernel; Step S2.2.1.4: Replace the mean and variance of all source domain datasets in the model with the mean and variance of the target domain dataset using adaptive batch normalization ;​​​ Step S2.2.1.5: Calculating the maximum mean difference between the source domain and the target domain datasets using a multi-kernel maximum average distance, and then combining it with the classification loss, the formula of the multi-kernel maximum average distance being: ， wherein: is a source domain dataset; is a target domain dataset; is an i-th source domain data in the source domain dataset; is a j-th target domain data in the target domain dataset; is a number of source domain datasets; is a number of target domain datasets; is a reproducing kernel Hilbert space; is a Gaussian kernel function; The loss function is calculated as follows: ， ， wherein, is a loss function of the multi-layer adaptive convolutional neural network, is a cross-entropy loss of the source domain sample, is a training sample, is a true label of the target domain sample, is a diagnosis result, is a parameter set, is a weight parameter; Step S2.2.1.6: Classifying the extracted features through a fully connected layer to realize the running state diagnosis of the threshing and separating device (1) and the cleaning device (2).

3. The real-time state diagnosis and regulation method of a migration learning-based de- cleaning device according to claim 1, characterized in that, The extraneous matter flow is preprocessed, specifically: The shaking plate (2-1) surface array installs m*n sensors, collects m*n exuviae flow signals, respectively as follows: 11 , w 12 , ……w 1n , ……w m1 , ……w mn ; Taking 1s as the time sequence segment length, the segments overlap by 50%, each extraneous matter flow signal is sliced, and then an (m x n) x 1 matrix is formed, so that n seconds of extraneous matter flow signal can obtain 2n-1 matrix extraneous matter flow signals; The 2n-1 matrix extraneous matter flow signals are filtered to eliminate the influence of vibration noise.

4. The real-time state diagnosis and regulation method of a migration learning-based de- cleaning device according to claim 1, characterized in that, The running state of the threshing and separating device (1) is combined with the torque of the threshing cylinder to adjust the rotation speed of the threshing cylinder and the concave clearance, specifically: When the running state of the threshing and separating device (1) is normal, if the torque of the threshing cylinder is less than the normal value, the feeding amount is small at this time, the threshing cylinder is in a low load state, and the rotation speed of the threshing cylinder needs to be reduced and the concave clearance needs to be reduced; When the running state of the threshing and separating device (1) is normal, if the torque of the threshing cylinder is greater than the normal value, the feeding amount is large at this time, the threshing cylinder is in a high load state, and the rotation speed of the threshing cylinder needs to be increased and the concave clearance needs to be increased; When the running state of the threshing and separating device (1) is abnormal, if the torque of the threshing cylinder is greater than the normal value and is in a continuously increasing state, it is in a clogging state at this time, and the rotation speed of the threshing cylinder needs to be increased and the concave clearance needs to be increased; When the running state of the threshing and separating device (1) is abnormal, if the torque of the threshing cylinder is greater than the normal value and is stable, it is in a clogging state at this time, and an alarm needs to be given and the work needs to be stopped.

5. The real-time condition diagnosis and regulation method of a migration learning-based de-stoning device according to claim 1, characterized in that, The running state of the cleaning device (2) is combined with the grain flow under the cleaning screen to adjust the frequency or amplitude of the shaking plate (2-1), specifically: When the running state of the cleaning device (2) is normal, the frequency or amplitude of the shaking plate (2-1) remains unchanged; When the running state of the cleaning device (2) is abnormal, if the grain flow under the cleaning screen is less than or equal to the normal range, the frequency or amplitude of the shaking plate (2-1) is increased; When the running state of the cleaning device (2) is abnormal, if the grain flow under the cleaning screen decreases rapidly from the normal range, it is in a clogging state at this time, the frequency or amplitude of the shaking plate (2-1) is increased, the frequency or amplitude of the fish scale screen (2-2) is increased, and the opening of the fish scale screen (2-2) is increased; When the running state of the cleaning device (2) is abnormal, if the grain flow under the cleaning screen is zero, it is in a clogging state at this time, and an alarm needs to be given and the work needs to be stopped.

6. The system of real-time state diagnosis and regulation of a migration learning-based destoning and scoping device according to any one of claims 1-5, characterized in that, It comprises a threshing and separating device (1), a cleaning device (2), an information monitoring system (3), an embedded processor (4) and a control system (5). The threshing and separating device (1) is provided with the cleaning device (2) at the bottom; the threshing and separating device (1) comprises a concave screen (1-1) with adjustable gap; the cleaning device (2) comprises a shaking plate (2-1), a fish scale screen (2-2), a woven screen (2-3), a grain flow sensor (2-4) and a vibration generator; the shaking plate (2-1) and the fish scale screen (2-2) are respectively installed at both sides of the outlet of the threshing and separating device (1), and the shaking plate (2-1) is located above the fish scale screen (2-2); the shaking plate (2-1) and the fish scale screen (2-2) are connected with the vibration generator respectively; the fish scale screen (2-2) is provided with the woven screen (2-3) at the bottom; the shaking plate (2-1) is provided with a sensor inside for monitoring the threshing material flow falling into the shaking plate (2-1); the grain flow sensor (2-4) is located at the bottom of the woven screen (2-3) for measuring the grain flow under the cleaning screen. The information monitoring system (3) comprises at least one sensor for monitoring the torque and rotating speed of the threshing cylinder; the information monitoring system (3) acquires the grain flow under the cleaning screen and the threshing material flow; the information monitoring system (3) inputs the collected and monitored information into the embedded processor (4) for diagnosing the state of the threshing and separating device (1) and the cleaning device (2); the control system (5) adjusts the rotating speed of the threshing cylinder, the gap of the concave screen and the amplitude or frequency of the shaking plate (2-1) according to the state of the threshing and separating device (1) and the cleaning device (2).

7. The system for real-time condition diagnosis and regulation of a device for cleaning by size and by material according to claim 6, characterized in that, The shaking plate (2-1) comprises an array shaking plate (2-1-1), a second shaking plate bracket (2-1-2), a sliding block (2-1-3), a base (2-1-4), a first damping (2-1-5) and a spoke (2-1-6); one end of the array shaking plate (2-1-1) is provided with the spoke (2-1-6) extending above the fish scale screen (2-2); the array shaking plate (2-1-1) is installed on the second shaking plate bracket (2-1-2) through the first damping (2-1-5); the second shaking plate bracket (2-1-2) is installed on the sliding block (2-1-3), which is movably installed in the sliding rail with slope of the base (2-1-4); the sliding block (2-1-3) reciprocates in the sliding rail with slope through an actuator, so that the second shaking plate bracket (2-1-2) reciprocates in height for changing the amplitude or frequency of the array shaking plate (2-1-1); the surface of the array shaking plate (2-1-1) is provided with a sensor for monitoring the threshing material flow falling into the array shaking plate (2-1-1).

8. The system for real-time condition diagnosis and regulation of a device for cleaning by size and by material according to claim 7, characterized in that, The array type shaking plate (2-1-1) comprises a shaking top plate (2-1-1-1), a piezoelectric sensor (2-1-1-2), a second shock absorption damping (2-1-1-4) and a first shaking plate support (2-1-1-3); a plurality of shaking top plates (2-1-1-1) are arrayed above the first shaking plate support (2-1-1-3), each shaking top plate (2-1-1-1) is connected with the first shaking plate support (2-1-1-3) through the second shock absorption damping (2-1-1-4); a piezoelectric sensor (2-1-1-2) is installed at the center of the bottom surface of each shaking top plate (2-1-1-1) for monitoring the flow of the detached material falling on the shaking top plate (2-1-1-1).

9. The system for real-time condition diagnosis and regulation of a device for decontaminated and cleaned separation based on transfer learning according to claim 6, characterized in that, The control system (5) comprises a threshing cylinder rotating speed controller, a concave plate gap controller and a shaking plate moving controller; the threshing cylinder rotating speed controller adjusts the rotating speed of the threshing cylinder according to the states of the threshing and separating device (1) and the cleaning device (2); the concave plate gap controller adjusts the concave plate gap according to the states of the threshing and separating device (1) and the cleaning device (2); and the shaking plate moving controller adjusts the moving speed and the reciprocating frequency of the sliding block according to the states of the threshing and separating device (1) and the cleaning device (2).

Images