A method for abnormal shutdown protection of a feeding assembly based on oil state sensing

Abstract

The application discloses a kind of based on oil state perception's feed assembly abnormal shutdown protection method, belong to equipment intelligent protection technical field, comprising: acquisition oil state parameter and the operating condition parameter of feed assembly;Through wavelet packet decomposition, short-time fourier transform is extracted feature, constructs double-channel deep confidence network model, and outputs the abnormal feature determination data containing abnormal type and severity;Adopt Monte Carlo simulation and particle swarm optimization algorithm, adaptive iteration correction grading shutdown protection threshold value;Real-time monitoring parameter and grading execution protection action, collect transient and steady-state data after shutdown and form lockout determination data;Combining oil state qualified interval and mechanical impact attenuation determines recovery state, not recovery then maintain lockout, recovery then release lockout and restart monitoring;The application realizes oil and operating condition coupling health assessment, threshold adaptive optimization and grading intelligent shutdown protection, improves the operation safety and reliability of feed assembly.

Description

Technical Field

[0001] This invention belongs to the field of intelligent equipment protection technology, specifically a method for abnormal shutdown protection of material conveying components based on oil status sensing. Background Technology

[0002] As the core actuators of industrial conveying systems, material conveying components often employ hydraulic and mechanical transmission structures. Operating under heavy loads and continuous conditions for extended periods, they are prone to wear, leakage, and jamming due to issues such as oil contamination, excessive moisture, and viscosity degradation. In severe cases, this can lead to unplanned equipment shutdowns and production line interruptions. Existing protection methods often rely on monitoring single operating parameters, failing to couple oil conditions with component operating conditions for evaluation. This results in insufficient accuracy in anomaly identification, leading to missed detections or false alarms. Traditional shutdown protection thresholds are often fixed, unable to adapt to dynamic changes in different operating conditions and oil degradation processes, resulting in a high false trigger rate and impacting production continuity. Furthermore, the lack of comprehensive interlocking judgment regarding oil conditions and mechanical impact after shutdown makes blind restarting prone to secondary damage. While some solutions incorporate oil detection, their feature extraction is limited and model fusion is low, making it difficult to accurately characterize early, subtle anomalies. Furthermore, they fail to retain transient critical data during emergency shutdowns, making fault tracing difficult. In the unrecovered state, there is a lack of periodic re-inspection mechanisms, compromising restart safety. In summary, existing technologies struggle to achieve full-cycle, adaptive, and highly reliable abnormal shutdown protection for material conveying components, failing to meet the demands for intelligent and safe operation in industrial settings. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a method for abnormal shutdown protection of material conveying components based on oil condition sensing. This method collects oil condition parameters and operating parameters of the material conveying component; extracts features through wavelet packet decomposition and short-time Fourier transform, constructs a dual-channel deep belief network model, and outputs anomaly feature judgment data containing anomaly type and severity; employs Monte Carlo simulation and particle swarm optimization algorithms to adaptively and iteratively correct the graded shutdown protection threshold; monitors parameters in real time and executes protection actions in stages; after shutdown, transient and steady-state data are collected to form lockout judgment data; the recovery status is determined by combining the acceptable oil condition range and mechanical impact attenuation; if recovery is not achieved, the lockout is maintained; if recovery is achieved, the lockout is released and monitoring is restarted. This invention achieves coupled health assessment of oil and operating conditions, adaptive threshold optimization, and graded intelligent shutdown protection, improving the operational safety and reliability of the material conveying component.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for abnormal shutdown protection of a material conveying assembly based on oil state sensing includes:

[0006] S1: Collect the oil state parameters in the working circuit of the material conveying component and the operating condition parameters of the material conveying component;

[0007] S2: Based on the oil state parameters and the operating condition parameters, construct a dual-channel deep confidence network model and output abnormal feature judgment data;

[0008] S3: Based on the abnormal feature determination data, perform adaptive iterative correction of the graded shutdown protection threshold until the false trigger probability of triggering shutdown action based on the obtained optimal graded shutdown protection threshold is lower than the preset probability threshold, and output the protection strategy ready signal.

[0009] S4: Based on the protection strategy ready signal, monitor the oil status parameters and operating condition parameters in real time. When the monitored oil status parameters and operating condition parameters reach the level shutdown condition corresponding to the optimal level shutdown protection threshold, trigger the corresponding shutdown action and collect the lockout judgment data after shutdown.

[0010] S5: Based on the lockout determination data, determine whether the oil condition has recovered to the preset acceptable oil condition range. If it has not recovered, maintain the shutdown lockout state; if it has recovered, release the lockout and restart monitoring.

[0011] Specifically, the steps of S2 include:

[0012] Read the oil state parameters and the operating condition parameters of the conveying assembly; the oil state parameters include at least the oil dielectric constant, oil viscosity index, oil moisture content and oil particle concentration; the operating condition parameters include at least the assembly outlet pressure, assembly vibration acceleration, motor stator current RMS value and conveying assembly spindle speed;

[0013] For the time-series data of oil particle concentration in the oil state parameters, a three-level decomposition is performed using Daubechies fourth-order wavelet packet decomposition to extract the energy proportion of each frequency band node, forming a frequency domain energy feature vector of particle concentration; for the time-series data of oil viscosity index in the oil state parameters, the sliding window average value is calculated and the first-order backward difference is obtained to obtain the viscosity change trend feature; the oil dielectric constant, oil moisture content, particle concentration frequency domain energy feature vector, and viscosity change trend feature are horizontally concatenated to output the oil degradation feature vector.

[0014] A short-time Fourier transform is performed on the component vibration acceleration in the operating condition parameters to obtain the vibration time-frequency feature matrix. The component outlet pressure, effective value of motor stator current, and spindle speed of the material conveying component in the operating condition parameters are normalized and then superimposed on the vibration time-frequency feature matrix in the channel dimension to output the operating condition feature matrix.

[0015] Specifically, the steps of S2 further include:

[0016] A dual-channel deep belief network model is constructed. The first channel is a stacked structure of three restricted Boltzmann machines, which takes the oil degradation feature vector as input and outputs the first deep feature vector. The second channel is a time-frequency conversion network consisting of two convolutional layers followed by two pooling layers, which takes the operating condition feature matrix as input and outputs the second deep feature vector. The first deep feature vector and the second deep feature vector are concatenated in a feature concatenation layer to output a joint feature vector.

[0017] The joint feature vector is sequentially input into two fully connected layers and the output layer to calculate the health index value at the current time. The health index value is compared with the preset health baseline threshold. When the health index value is lower than the health baseline threshold and the difference exceeds the preset deviation threshold, abnormal feature judgment data is output. The abnormal feature judgment data includes an abnormality type label and an abnormality severity level.

[0018] Specifically, the steps of S3 include:

[0019] Read the pre-stored set of graded shutdown protection thresholds; the set of graded shutdown protection thresholds includes a first-level early warning threshold, a second-level load reduction threshold, and a third-level emergency shutdown threshold;

[0020] The system analyzes the severity level of anomalies in the data and assigns initial adjustment coefficients to the first-level warning threshold, the second-level load reduction threshold, and the third-level emergency shutdown threshold according to the pre-stored level threshold mapping table. The system then outputs the set of initially adjusted level shutdown protection thresholds.

[0021] The initial adjusted set of hierarchical shutdown protection thresholds is used as the current threshold. The historical abnormal feature judgment data sequence obtained by cumulative storage is used as the sample and input into the Monte Carlo simulator. Each time, a sample is randomly selected and a shutdown trigger decision is made according to the current threshold. The total number of shutdowns triggered and the number of false triggers are counted to calculate the current false trigger probability.

[0022] Specifically, the steps of S3 further include:

[0023] The current false trigger probability is compared with a preset probability threshold. If the current false trigger probability is greater than the preset probability threshold, the particle swarm optimization algorithm is started: the first-level warning threshold, the second-level load reduction threshold, and the third-level emergency shutdown threshold are used as variables to be optimized, and the current false trigger probability is used as the fitness function. The optimal set of graded shutdown protection thresholds is searched to minimize the false trigger probability. The current threshold is replaced with the optimal set of graded shutdown protection thresholds, and the iteration is repeated until the current false trigger probability is lower than the preset probability threshold.

[0024] When the iteration stops, the final optimal set of hierarchical shutdown protection thresholds is written into non-volatile memory, and incremental logic is performed on the first-level warning threshold, the second-level load reduction threshold, and the third-level emergency shutdown threshold to ensure that the first-level warning threshold is less than the second-level load reduction threshold and the second-level load reduction threshold is less than the third-level emergency shutdown threshold. After the incremental logic verification passes, a protection strategy ready signal containing the system time and threshold version number is generated.

[0025] Specifically, the specific steps of S4 include:

[0026] Based on the protection strategy ready signal, the oil status parameters and operating condition parameters are monitored in real time. When any of the monitored parameters reaches the level shutdown condition corresponding to the optimal level shutdown protection threshold, the corresponding shutdown action is executed according to the determined level. Among them, triggering the first level early warning only outputs an alarm signal, triggering the second level reduces the motor output power, and triggering the third level emergency shutdown sends a hard wire cut-off signal to the motor drive circuit.

[0027] While executing a level 3 emergency shutdown and sending a hard-wire cut-off signal, a power supply maintenance window with a preset delay duration is initiated, and the power supply maintenance window is divided into an initial period and a middle period.

[0028] Specifically, the steps of S4 further include:

[0029] Within the power supply window, power is continuously supplied to the sensors corresponding to the oil state parameters and operating condition parameters. During the initial period of the power supply window, the oil particle concentration and component vibration acceleration are collected at a preset sampling interval. The maximum values ​​are taken as the peak values ​​of the transient oil particle concentration and the transient vibration acceleration during shutdown. At the same time, the oil dielectric constant is collected and the maximum value is taken as the peak value of the transient dielectric constant during shutdown.

[0030] During the middle period of the power supply window, the oil viscosity index and oil moisture content were continuously sampled and the arithmetic mean was calculated, which were used as the steady-state values ​​of the shutdown viscosity index and the shutdown moisture content, respectively.

[0031] A unique shutdown event number is generated for this shutdown action. The peak values ​​of transient oil particle concentration, transient vibration acceleration, and transient dielectric constant during shutdown, as well as the steady-state values ​​of viscosity index and moisture content during shutdown, are structured and encapsulated, and a latch-up judgment data packet is output.

[0032] Specifically, the specific steps of S5 include:

[0033] The steady-state value of shutdown moisture content, the peak value of shutdown transient oil particle concentration, the peak value of shutdown transient dielectric constant, and the steady-state value of shutdown viscosity index are extracted from the lockout judgment data packet. They are compared with the preset upper and lower thresholds of moisture content, particle concentration, dielectric constant, and viscosity index in the qualified oil state range. When the steady-state value of shutdown moisture content is lower than the upper threshold of moisture content, the peak value of shutdown transient oil particle concentration is lower than the upper threshold of particle concentration, the peak value of shutdown transient dielectric constant is between the upper and lower thresholds of dielectric constant, and the steady-state value of shutdown viscosity index is between the upper and lower thresholds of viscosity index, the basic state of oil is qualified. Otherwise, the judgment result that the oil state has not been restored is output and the shutdown lockout state is maintained.

[0034] When the output signal indicating that the basic state of the oil is qualified, the component vibration acceleration time-series sampling sequence that is from the same source as the peak value of the transient vibration acceleration during shutdown is extracted from the lockout judgment data packet. The component vibration acceleration time-series sampling sequence is subjected to fast Fourier transform to obtain the frequency spectrum. The ratio of the sum of the energy amplitudes corresponding to the preset gear meshing frequency and its harmonics in the frequency spectrum to the total energy amplitude of the entire frequency band is calculated to obtain the main frequency energy ratio. When the main frequency energy ratio is lower than the preset background noise energy ratio threshold, the mechanical shock attenuation signal is output. Otherwise, the judgment result that the oil state has not been restored is output and the shutdown lockout state is maintained.

[0035] When the output oil condition is qualified and the mechanical shock has decayed, the output oil condition has been restored. The shutdown lockout state is released and the system returns to S4 to restart real-time monitoring. Otherwise, the shutdown lockout state is maintained.

[0036] Specifically, after determining that the output oil state has not recovered, a retest based on periodic idling is also included, specifically:

[0037] After outputting the judgment result that the oil condition has not been restored and maintaining the shutdown lockout state, the periodic re-inspection timer is started. The preset duration of the periodic re-inspection timer is the waiting period for the re-inspection test.

[0038] When the periodic re-inspection timer expires, the pressure relief bypass valve pre-installed at the oil inlet of the conveying component is opened, so that the oil circuit forms a low-pressure circulation loop. The motor is driven by the frequency converter to run at a preset percentage of the rated speed as the idling speed for a preset cleaning time, and the oil particle concentration, oil moisture content and component outlet pressure are collected again to form re-inspection lockout judgment data. The oil particle concentration, oil moisture content and component outlet pressure are used to judge the oil cleanliness, dryness and oil circuit fullness, respectively.

[0039] Specifically, after determining that the output oil state has not recovered, a retest based on periodic idling is also included, which specifically includes:

[0040] The oil particle concentration in the output re-inspection lockout judgment data is compared with the upper limit threshold of particle concentration in the qualified oil condition range. The water content in the re-inspection lockout judgment data is compared with the upper limit threshold of water content in the qualified oil condition range. The component outlet pressure in the re-inspection lockout judgment data is compared with the preset minimum start-up outlet pressure threshold.

[0041] When the oil particle concentration is lower than the upper limit threshold for particle concentration, the oil moisture content is lower than the upper limit threshold for moisture content, and the component outlet pressure is higher than the minimum start-up outlet pressure threshold, the result indicating that the oil condition has been restored during the re-inspection is output; otherwise, the periodic re-inspection timer is reset, and the system waits for the next re-inspection test.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] 1. This invention achieves accurate identification of early subtle anomalies in the material conveying component by coupling monitoring of oil state and operating conditions in two dimensions, combined with wavelet packet decomposition, short-time Fourier transform and dual-channel deep belief network model. It can output judgment results including anomaly type and severity, improve the accuracy and sensitivity of fault detection, and effectively solve the problem of easy omission and misjudgment in traditional single parameter monitoring.

[0044] 2. This invention uses Monte Carlo simulation and particle swarm optimization algorithm to adaptively and iteratively correct the graded shutdown protection threshold, which can dynamically match the working conditions and oil deterioration state, control the probability of false shutdown within a preset range, and at the same time, through the graded protection strategy of first-level early warning, second-level load reduction and third-level emergency shutdown, it takes into account both equipment safety and production continuity, and avoids the defects of fixed threshold protection schemes such as poor flexibility and unreasonable protection actions.

[0045] 3. This invention collects transient and steady-state data after shutdown to form the basis for lockout determination. It combines the qualified oil range and mechanical impact attenuation to comprehensively determine the restart conditions. If the system does not recover, the lockout is maintained and periodic low-speed idling re-inspection is initiated, which can effectively prevent secondary damage caused by blind restart. It also fully retains key shutdown data, which facilitates fault tracing and maintenance decisions, and greatly improves the intelligence level and operational reliability of the material conveying component protection. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of an abnormal shutdown protection method for a material conveying assembly based on oil state sensing according to the present invention;

[0047] Figure 2 This is a flowchart illustrating the principle of an abnormal shutdown protection method for a material conveying assembly based on oil state sensing, according to the present invention. Detailed Implementation

[0048] Please see Figure 1and Figure 2 The present invention provides an embodiment of a method for abnormal shutdown protection of a material conveying assembly based on oil state sensing. In this embodiment, the material conveying assembly refers to a gear pump. The method includes steps S1 to S5, namely:

[0049] S1: Collect the oil state parameters in the working circuit of the material conveying component and the operating condition parameters of the material conveying component;

[0050] Furthermore, multiple types of integrated sensors are installed on the oil inlet and outlet pipes of the material conveying assembly, the bottom of the oil tank, and the spindle bearing housing. All sensors are connected to the edge computing control unit via an industrial fieldbus. The sensors are powered by DC isolation to avoid power supply interference affecting the accuracy of the acquired signals. The oil status parameter acquisition sensor adopts a non-contact installation structure, which is close to the outer wall of the pipe or directly immersed in the oil inside the oil tank, without damaging the sealing and flow of the oil circuit. The operating condition parameter acquisition sensor adopts a rigid fixed installation method, which keeps it in close contact with the housing of the material conveying assembly, the motor housing, and the end of the spindle to ensure that the physical signals during the equipment operation can be accurately transmitted.

[0051] Furthermore, all sensors employ a high-precision time synchronization mechanism based on the IEEE 1588v2 protocol to synchronize the local acquisition clock of each sensor to microsecond-level accuracy. This ensures that the oil state parameters and operating condition parameters are strictly aligned in the time dimension, preventing timing misalignment of subsequent model input data due to time asynchrony. During normal operation, the sensors continuously acquire data at a fixed sampling frequency. The sampling frequency for oil state parameters is set to 10Hz, and the sampling frequency for operating condition parameters is set to 50Hz. For parameters with rapid transient changes, such as vibration and current, the sampling frequency is increased to 100Hz to ensure complete capture of the dynamic changes in the parameters.

[0052] Furthermore, the collected raw data undergoes preliminary filtering in the sensor's built-in preprocessing unit to remove useless signals such as power frequency interference and random noise, retaining the true and valid feature signals. Subsequently, the processed data is encapsulated into a data packet containing the sensor number, geographic location tag, timestamp, and parameter values, and transmitted in real time to the edge computing control unit via the fieldbus. The edge computing control unit performs timestamp verification and integrity verification on the received data packet, eliminating invalid data with abnormal timestamps, missing data, or overflowing values, and sorting and storing the valid data in chronological order to form a continuous, stable, and time-synchronized sequence of oil state parameters and operating condition parameters.

[0053] Furthermore, this embodiment uses a material conveying assembly system operating continuously in an industrial workshop as a typical application scenario. This material conveying assembly operates continuously for 24 hours a day, and the working environment includes interference factors such as dust, temperature fluctuations, and mechanical vibration. An oil dielectric constant sensor, an oil viscosity index sensor, an oil moisture content sensor, and an oil particle concentration sensor are installed at the oil main outlet of the system. A pressure sensor is installed at the outlet end of the material conveying assembly, a vibration acceleration sensor is installed on the outside of the housing, a current acquisition module is installed in the drive motor junction box, and a speed sensor is installed at the end of the spindle. All sensors synchronously collect data, forming a multi-dimensional parameter sequence including timestamps and location identifiers, fully reflecting the real-time changes in the oil state of the material conveying assembly's working circuit and the equipment's own operating status.

[0054] S2: Based on the oil state parameters and the operating condition parameters, construct a dual-channel deep confidence network model and output abnormal feature judgment data;

[0055] S3: Based on the abnormal feature determination data, perform adaptive iterative correction of the graded shutdown protection threshold until the false trigger probability of triggering shutdown action based on the obtained optimal graded shutdown protection threshold is lower than the preset probability threshold, and output the protection strategy ready signal.

[0056] S4: Based on the protection strategy ready signal, monitor the oil status parameters and operating condition parameters in real time. When the monitored oil status parameters and operating condition parameters reach the level shutdown condition corresponding to the optimal level shutdown protection threshold, trigger the corresponding shutdown action and collect the lockout judgment data after shutdown.

[0057] S5: Based on the lockout determination data, determine whether the oil condition has recovered to the preset acceptable oil condition range. If it has not recovered, maintain the shutdown lockout state; if it has recovered, release the lockout and restart monitoring.

[0058] The specific steps of S2 include:

[0059] S2.1: Read the oil state parameters and the operating condition parameters of the conveying component; the oil state parameters include at least the oil dielectric constant, oil viscosity index, oil moisture content and oil particle concentration; the operating condition parameters include at least the component outlet pressure, component vibration acceleration, motor stator current RMS value and conveying component spindle speed;

[0060] Furthermore, the dielectric constant of the oil reflects the overall aging, contamination, and water content of the oil. When water is mixed into the oil, oxidation products are generated, and impurities increase, the dielectric constant will shift significantly, making it a core indicator for judging the overall performance of the oil. The viscosity index of the oil directly determines its lubrication and fluid properties. An abnormal viscosity index can lead to insufficient lubrication, increased flow resistance, and accelerated component wear. The water content of the oil is a key indicator of oil contamination. Water can damage the oil film structure, causing corrosion and emulsification, and accelerating the failure of internal components. The oil particle concentration reflects the amount of metal abrasive particles, dust, and impurities in the oil. Excessive particle concentration indicates abnormal wear inside the component and is an important signal of early failure.

[0061] Furthermore, the component outlet pressure reflects the load status and oil circuit unobstructedness of the material conveying component. Abnormally high pressure may indicate oil circuit blockage or excessive load, while low pressure may indicate leakage or cavitation failure. The component vibration acceleration contains fault information such as equipment friction, wear, loosening, and impact, and is a direct basis for judging the mechanical condition. The effective value of the motor stator current can reflect the load changes and electrical faults of the drive system. Sudden increases or decreases in current correspond to different types of abnormalities. The spindle speed of the material conveying component directly reflects the operating speed of the equipment. Speed ​​fluctuations and stalling are intuitive manifestations of equipment abnormalities.

[0062] S2.2: For the time-series data of oil particle concentration in the oil state parameters, a three-level decomposition is performed using Daubechies fourth-order wavelet packet decomposition to extract the energy proportion of each frequency band node, forming a frequency domain energy feature vector of particle concentration; for the time-series data of oil viscosity index in the oil state parameters, the sliding window average value is calculated and the first-order backward difference is obtained to obtain the viscosity change trend feature; the oil dielectric constant, oil moisture content, particle concentration frequency domain energy feature vector, and viscosity change trend feature are horizontally concatenated to output the oil degradation feature vector;

[0063] It should be noted that the time series data of oil particle concentration contains a large number of frequency domain features. Particles of different sizes and from different sources will exhibit specific energy distributions in different frequency bands, which cannot be accurately distinguished by ordinary time domain analysis. However, by using Daubechies fourth-order wavelet packets for three-level decomposition, the original particle concentration signal can be uniformly divided in the entire frequency band, resulting in node signals in 8 different frequency bands. Each frequency band node corresponds to the response range of different particle characteristics.

[0064] Furthermore, the time-series data of oil particle concentration in the oil state parameters is decomposed into three levels using Daubechies fourth-order wavelet packet decomposition to extract the energy proportion of each frequency band node, including:

[0065] (1) Select 1000 consecutive points of oil particle concentration time series data in the oil state parameters, and arrange the data points in chronological order;

[0066] (2) The fourth-order Daubechies wavelet basis is selected as the basis function for wavelet packet decomposition. Three-level wavelet packet decomposition is performed on the time series data of oil particle concentration. The three-level decomposition is strictly performed according to the rule of refining the frequency band layer by layer. The first level of decomposition divides the original oil particle concentration time series data into two frequency band nodes. The second level of decomposition further divides each frequency band of the previous level into two sub-frequency bands. The third level of decomposition further subdivides all frequency bands of the previous level. After the complete three-level decomposition, eight independent frequency band nodes are finally obtained. Each frequency band node corresponds to a signal component in a fixed frequency range.

[0067] (3) After the decomposition is completed, the energy of the signal corresponding to each frequency band node is calculated. The calculation method is to square the values ​​of all data points in the frequency band node and then sum them to obtain the single energy value of the frequency band. The single energy values ​​of the 8 frequency band nodes are added together to obtain the total energy of all frequency bands after the oil particle concentration time series data is decomposed.

[0068] (4) Divide the single energy value of each frequency band node by the total energy of all frequency bands to obtain the energy ratio of the frequency band node. The energy ratios of the 8 frequency band nodes are arranged in the order of the decomposed frequency bands to form a complete particle concentration frequency domain energy feature vector, which can comprehensively characterize the energy distribution of oil particle concentration in different frequency ranges.

[0069] Furthermore, for the time-series data of the oil viscosity index in the oil state parameters, after calculating the sliding window average value, the first-order backward difference is obtained to obtain the viscosity change trend characteristics, including:

[0070] (1) Processing the time series data of oil viscosity index in the oil state parameters, including: setting the length of the sliding window. In this embodiment, the length of the sliding window is set to 50 consecutive time series data points, and the sliding step size is set to 1 data point to ensure that the window slides without gaps and covers all data.

[0071] (2) The time series data of oil viscosity index is extracted sequentially according to the set sliding window. For each window of data extracted, the arithmetic mean of all oil viscosity index values ​​in that window is calculated. As the window slides point by point, a continuous sliding window average sequence is generated. The sliding window average sequence can eliminate the interference of instantaneous fluctuations and highlight the overall trend of oil viscosity index.

[0072] (3) After obtaining the sliding window average value sequence, perform first-order backward difference calculation. The difference calculation rule is to subtract the sliding window average value of the previous position from the current position's sliding window average value, and complete the difference operation of all positions point by point. All the difference results are combined together to form the viscosity change trend characteristics, which can intuitively reflect the rising and falling state and the speed of change of the oil viscosity index.

[0073] S2.3: Perform a short-time Fourier transform on the component vibration acceleration in the operating condition parameters to obtain a vibration time-frequency feature matrix. Normalize the component outlet pressure, effective value of motor stator current, and main shaft speed of the conveying component in the operating condition parameters and superimpose them with the vibration time-frequency feature matrix in the channel dimension to output the operating condition feature matrix. The short-time Fourier transform and normalization processing are existing technologies in this field and are not inventive solutions of this application, so they will not be described in detail here.

[0074] It should be noted that component vibration acceleration is a rapidly changing transient signal containing a large amount of non-stationary fault information. Simple time-domain analysis can only obtain the vibration amplitude and cannot pinpoint the frequency and time of the fault occurrence. Simple frequency-domain analysis cannot reflect the change process of the fault over time. Short-time Fourier transform can analyze vibration signals simultaneously in both time and frequency dimensions, converting the one-dimensional vibration acceleration time-series signal into a two-dimensional time-frequency feature matrix. The rows of the matrix represent the time dimension, and the columns represent the frequency dimension. The values ​​in the time-frequency feature matrix correspond to the vibration energy at a specific time and frequency, completely preserving the time, frequency, and amplitude information of the vibration signal, and can accurately identify fault characteristics such as vibration impact, frequency shift, and abnormal resonance.

[0075] S2.4: Construct a dual-channel deep belief network model; the first channel is a three-layer restricted Boltzmann machine stacked structure, which takes the oil degradation feature vector as input and outputs the first deep feature vector; the second channel is a time-frequency conversion network with two convolutional layers followed by two pooling layers, which takes the operating condition feature matrix as input and outputs the second deep feature vector; the first deep feature vector and the second deep feature vector are concatenated in the feature concatenation layer to output the joint feature vector;

[0076] Furthermore, the first channel employs a three-layer Restricted Boltzmann Machine (RBM) stacked structure to process one-dimensional oil degradation feature vectors. The input layer of the first RBM has 128 neurons, matching the dimension of the oil degradation feature vector, and 90 neurons in the hidden layer. Inter-layer connection weights are initialized using a random normal distribution, and all bias parameters are initialized to zero. This layer is responsible for preliminary feature extraction and nonlinear mapping of the original input oil degradation features, preserving the basic distribution information of the features. The second RBM uses the hidden layer output of the first RBM as input, with 90 neurons in the input layer and 64 neurons in the hidden layer. Weights and bias parameters follow the same initialization rules. This layer is responsible for further fusing the basic features extracted by the first layer, uncovering the relationships between features, weakening redundant information, and strengthening effective features. The third RBM... The Boltzmann machine takes the output of the hidden layer of the second layer of the Restricted Boltzmann Machine as input, with 64 neurons in the input layer and 32 neurons in the hidden layer. This layer is the final feature output layer of the first channel, responsible for highly abstracting and compressing the features extracted from the first two layers to form deep features with reduced dimensions and strong representational capabilities. The three-layer Restricted Boltzmann Machine adopts a layer-by-layer unsupervised training method. First, the first layer is trained alone. After completion, the parameters of the first layer are fixed. Then, the second layer is trained. After completion, the parameters of the second layer are fixed. Finally, the third layer is trained. During training, the learning rate is set to 0.001, the number of iterations per training session is set to 200, and the number of divergence steps is set to 1. After all training is completed, the oil degradation feature vector is input into the first channel and calculated by the three-layer Restricted Boltzmann Machine through layer-by-layer forward propagation. Finally, the first deep feature vector with fixed dimensions is output. The Restricted Boltzmann Machine and the random normal distribution are existing technologies in this field and are not inventive solutions of this application. They will not be described in detail here.

[0077] Furthermore, the second channel is a time-frequency transformation network consisting of two convolutional layers followed by two pooling layers, used to process the two-dimensional operating condition feature matrix. First, the input size of the operating condition feature matrix is ​​determined. Based on the actual acquired and processed specifications, the height of the input matrix is ​​set to 64, the width to 64, and the number of channels to 4. This size perfectly matches the output specifications after processing the operating condition parameters. The first convolutional layer uses 3×3 convolutional units, with 16 kernels, a stride of 1, and edge padding to maintain the same padding size as the input. A non-linear activation function is used. This layer extracts local time-frequency correlation features and operating condition coupling features from the operating condition feature matrix. The first pooling layer immediately follows the first convolutional layer, using max pooling with a 2×2 kernel and a stride of 2. This layer downsamples the feature map output by the first convolutional layer. The first layer retains core features while reducing feature data volume and improving computational efficiency. The second convolutional layer takes the output feature map of the first pooling layer as input, uses 3×3 convolutional units with 32 kernels, a stride of 1, and the same edge padding method. The activation function remains consistent. This layer is used to further extract deeper time-frequency distribution features and multi-parameter coupled features, enhancing the discriminative ability of the features. The second pooling layer follows the second convolutional layer, using the same max pooling method as the first pooling layer, with a 2×2 kernel and a stride of 2, to resample the features. The second channel is trained using supervised fine-tuning with a learning rate of 0.0005 and 150 iterations. After training, the running condition feature matrix is ​​input to the second channel, and after sequential calculation by two convolutional layers and two pooling layers, a fixed-dimensional second deep feature vector is finally output.

[0078] S2.5: Input the joint feature vector into two fully connected layers and the output layer in sequence to calculate the health index value at the current time; compare the health index value with the preset health baseline threshold; when the health index value is lower than the health baseline threshold and the difference exceeds the preset deviation threshold, output the abnormal feature judgment data; the abnormal feature judgment data includes the abnormal type label and the abnormal severity level.

[0079] It should be noted that the health baseline threshold is a standard value obtained by statistically analyzing a large amount of normal operation data of the material conveying components and oil in brand new condition. It represents the health index corresponding to the ideal health state of the system. The preset deviation threshold is a threshold set according to the normal fluctuation range and slight aging range allowed by the equipment. It is used to distinguish between normal fluctuation and abnormal degradation. In this embodiment, the health baseline threshold is set to 0.85 and the deviation threshold is set to 0.1.

[0080] Furthermore, the specific steps of S2.5 include:

[0081] (1) Construct the first fully connected layer. The first fully connected layer takes the joint feature vector as the direct input. Based on the dimension determined by the joint feature vector, the number of input neurons of the first fully connected layer is set to 64 and the number of output neurons is set to 32. The inter-layer connection weights are randomly initialized using the standard normal distribution. The initial values ​​of the bias terms are all set to zero. The activation function is a non-linear activation function. During the training phase, the learning rate of this layer is set to 0.001. The batch size of a single iteration is set to 16 samples, and the total number of iterations is set to 150. This layer mainly performs preliminary feature mapping and dimension compression on the joint feature vector, filters redundant information, and retains the feature components that are strongly correlated with the health status.

[0082] (2) Construct the second fully connected layer. The second fully connected layer takes the output of the first fully connected layer as input. The number of input neurons is set to 32, and the number of output neurons is set to 8. The number of output neurons is consistent with the preset number of health levels. The weight initialization method and the initial bias value are the same as the previous layer. The activation function continues to use the same non-linear activation function. The learning rate during the training phase is set to 0.0005. The number of iterations and the batch size are consistent with the previous layer. This layer further abstracts and integrates the features, transforming high-dimensional features into low-dimensional discriminative features that can be directly used to distinguish health status.

[0083] (3) After feature processing is completed in two fully connected layers, the softmax output layer is connected. This layer takes the output of the second fully connected layer as input, with 8 input neurons and 1 output neuron. This layer does not set any additional activation function or trainable parameters, but only performs softmax normalization calculation to map the input feature values ​​to a continuous range of values ​​from 0 to 1. The final output value is the health index value of the device and oil coupling system at the current moment, which directly represents the current health status level of the system.

[0084] (4) After the health index value is calculated, the health status determination stage is entered. First, the health baseline threshold stored in the system storage unit is called. The health baseline threshold is obtained by statistical analysis of the normal operation data of brand new equipment and brand new oil. It is fixed at 0.85, which represents the minimum health standard when the system is in an ideal health state. Then, the difference between the current health index value and the health baseline threshold is calculated. The calculation method is to subtract the current health index value from the health baseline threshold to obtain the absolute difference between the two.

[0085] (5) Simultaneously call the preset deviation threshold value. The deviation threshold value is determined according to the normal operating fluctuation range of the equipment and the natural aging rate of the oil. It is fixed at 0.10, which represents the maximum deviation range between the normal fluctuation and natural aging that the system is allowed to generate. Compare the calculated difference with the preset deviation threshold. When the health index value is less than the health baseline threshold and the difference between the two is greater than the preset deviation threshold, it is determined that the current system has exceeded the normal fluctuation range and a real abnormal state has occurred.

[0086] (6) After confirming the abnormal state, the system matches the preset abnormal type mapping rules according to the feature distribution information carried in the joint feature vector. The abnormal types include oil contamination abnormality, oil aging abnormality, mechanical wear abnormality, load fluctuation abnormality, electrical operation abnormality and oil circuit flow abnormality, etc. The system automatically assigns the corresponding abnormal type label according to the feature matching result. At the same time, the system divides the severity level of the abnormality according to the magnitude of the health index value below the health baseline threshold. The level division rule is fixed in advance. When the difference is greater than 0.10 and less than or equal to 0.20, it is judged as a mild abnormality. When the difference is greater than 0.20 and less than or equal to 0.30, it is judged as a moderate abnormality. When the difference is greater than 0.30, it is judged as a severe abnormality. The system assigns the corresponding abnormality severity level according to the calculation result.

[0087] Furthermore, the anomaly type mapping rule is constructed as follows: before the equipment is put into operation, oil degradation feature vectors and operating condition feature matrices under various single fault and compound fault conditions are collected and input into a dual-channel deep belief network model to complete supervised training. The joint feature distribution intervals corresponding to different fault categories are stored as feature templates to form a mapping rule that corresponds one-to-one between feature distribution and anomaly type. During operation, the joint feature vectors collected in real time are matched with each fault feature template to achieve automatic identification of anomaly type.

[0088] (7) Combine and encapsulate the determined anomaly type labels and anomaly severity levels to form complete anomaly feature judgment data, which can be directly used for adaptive iterative correction of the graded shutdown protection threshold.

[0089] The specific steps of S3 include:

[0090] S3.1: Read the pre-stored set of graded shutdown protection thresholds; the set of graded shutdown protection thresholds includes a first-level early warning threshold, a second-level load reduction threshold, and a third-level emergency shutdown threshold;

[0091] Furthermore, the set of graded shutdown protection thresholds are initial thresholds preset at the factory based on equipment design parameters, industry standards, and safety specifications. They are stored in non-volatile memory and will not be lost due to power outages. The first-level warning threshold corresponds to a minor abnormal state, requiring only an alert to the operator without the need for shutdown or load reduction. The second-level load reduction threshold corresponds to a moderate abnormal state, requiring a reduction in equipment load to slow the rate of abnormality deterioration. The third-level emergency shutdown threshold corresponds to a severe abnormal state, necessitating immediate power cut-off and forced shutdown to protect equipment and personnel safety. The three thresholds increase progressively according to the severity of the abnormality, forming a gradient protection logic that balances equipment protection and production continuity.

[0092] S3.2: Analyze the severity level of the abnormality in the abnormal feature judgment data, and assign initial adjustment coefficients to the first-level early warning threshold, the second-level load reduction threshold and the third-level emergency shutdown threshold according to the pre-stored level threshold mapping table, and output the set of level shutdown protection thresholds after initial adjustment;

[0093] Furthermore, the specific steps in S3.2 include:

[0094] (1) Extract the severity level of anomalies from the anomaly feature determination data;

[0095] (2) After the extraction and confirmation of the severity level of the anomaly is completed, the level threshold mapping table stored in the non-volatile memory inside the system is called. The level threshold mapping table is a fixed parameter correspondence table. It has been calibrated before leaving the factory and will not be modified arbitrarily during use. The level threshold mapping table clearly specifies the initial adjustment coefficients of the first-level warning threshold, the second-level load reduction threshold and the third-level emergency shutdown threshold corresponding to the three levels of mild anomaly, moderate anomaly and severe anomaly, respectively. All coefficients are fixed values ​​and there are no random values ​​or dynamic changes.

[0096] (3) When the severity level of the current abnormality is determined to be mild, the corresponding initial adjustment coefficient is retrieved from the level threshold mapping table, wherein the initial adjustment coefficient of the first-level warning threshold is set to 0.95, the initial adjustment coefficient of the second-level load reduction threshold is set to 1.05, and the initial adjustment coefficient of the third-level emergency shutdown threshold is set to 1.10.

[0097] (4) When the severity level of the current anomaly is determined to be moderate, the corresponding initial adjustment coefficient is retrieved from the level threshold mapping table, wherein the initial adjustment coefficient of the first-level warning threshold is set to 0.85, the initial adjustment coefficient of the second-level load reduction threshold is set to 0.95, and the initial adjustment coefficient of the third-level emergency shutdown threshold is set to 1.00.

[0098] (5) When the severity level of the current anomaly is determined to be severe, the corresponding initial adjustment coefficient is retrieved from the level threshold mapping table, wherein the initial adjustment coefficient of the first-level warning threshold is set to 0.75, the initial adjustment coefficient of the second-level load reduction threshold is set to 0.85, and the initial adjustment coefficient of the third-level emergency shutdown threshold is set to 0.95.

[0099] (6) After retrieving the three sets of initial adjustment coefficients that match the current severity level of the anomaly, read the original graded shutdown protection threshold set stored in the system. The original graded shutdown protection threshold set includes the factory default first-level early warning original threshold, second-level load reduction original threshold and third-level emergency shutdown original threshold. All original thresholds are fixed values ​​and are the basis for calculating the threshold after initial adjustment.

[0100] (7) Perform the initial adjustment calculations for the three sets of thresholds respectively. First, multiply the original threshold of the first-level warning by the corresponding initial adjustment coefficient of the first-level warning threshold to obtain the adjusted first-level warning threshold. Then, multiply the original threshold of the second-level load reduction by the corresponding initial adjustment coefficient of the second-level load reduction threshold to obtain the adjusted second-level load reduction threshold. Then, multiply the original threshold of the third-level emergency shutdown by the corresponding initial adjustment coefficient of the third-level emergency shutdown threshold to obtain the adjusted third-level emergency shutdown threshold.

[0101] (8) After completing the independent calculation of the three sets of thresholds, perform a simple numerical verification of the calculation results to confirm that the adjusted first-level warning threshold is less than the adjusted second-level load reduction threshold, and the adjusted second-level load reduction threshold is less than the adjusted third-level emergency shutdown threshold, so as to ensure that the three-level thresholds maintain a logical relationship of progressively increasing and that there is no numerical inversion or logical conflict.

[0102] (9) After the numerical verification is passed, the adjusted first-level early warning threshold, the adjusted second-level load reduction threshold and the adjusted third-level emergency shutdown threshold are combined in sequence to form a complete set of initially adjusted graded shutdown protection thresholds, which are used as the initial input thresholds for Monte Carlo simulation and particle swarm optimization algorithm.

[0103] S3.3: Using the initially adjusted set of hierarchical shutdown protection thresholds as the current threshold, and the historical abnormal feature judgment data sequence obtained from the cumulative storage as a sample, input it into the Monte Carlo simulator. Each time, randomly sample and make a shutdown trigger decision according to the current threshold. Count the total number of shutdowns triggered and the number of false triggers, and calculate the current false trigger probability.

[0104] Furthermore, the specific steps of S3.3 include:

[0105] (1) The set of hierarchical shutdown protection thresholds obtained after initial adjustment is used as the threshold data for current simulation judgment. The current thresholds include three items: first-level early warning threshold, second-level load reduction threshold and third-level emergency shutdown threshold. At the same time, the historical abnormal feature judgment data sequence accumulated in the system's local data storage unit is called. The historical abnormal feature judgment data sequence consists of all abnormal feature records that are actually collected and judged during the long-term operation of the equipment. The data sequence length is set to no less than 10,000 samples. The samples cover all scenarios such as normal operation state, real abnormal state, and instantaneous interference state to ensure the comprehensiveness and representativeness of the simulation samples.

[0106] (2) The current threshold and the historical anomaly feature judgment data sequence are synchronously input into the Monte Carlo simulator. Before the Monte Carlo simulator is started, the fixed parameters need to be configured. The total number of sampling times for a single simulation is set to 10,000. After the Monte Carlo simulator is started, a single sample is drawn from the historical anomaly feature judgment data sequence according to the uniform random sampling rule. Each sampling is completed independently and the same sample is not drawn repeatedly. The sampling process strictly follows the principle of random distribution to ensure that the probability of the sample being selected is equal. The Monte Carlo simulator is the prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0107] (3) After extracting a single sample, the Monte Carlo simulator executes a shutdown trigger decision according to the current threshold. The decision process compares the oil state parameters in the sample with the operating condition parameters and the first-level warning threshold, the second-level load reduction threshold, and the third-level emergency shutdown threshold in the current threshold one by one. When the parameter value in the sample meets the trigger condition of any first-level threshold, it is marked as a shutdown trigger behavior. When the parameter value in the sample does not meet the trigger condition of any first-level threshold, it is marked as a non-triggered behavior.

[0108] (4) During the judgment process, the distinction between real trigger and false trigger is completed simultaneously. The system pre-labels the real status of each historical abnormal feature judgment data sample. The real status labels are divided into two categories: real abnormal and normal fluctuation. After the judgment is completed, if the real status label corresponding to the sample marked as the shutdown trigger behavior is real abnormal, it is judged as a valid trigger. If the real status label corresponding to the sample marked as the shutdown trigger behavior is normal fluctuation, it is judged as a false trigger.

[0109] (5) After completing a single sampling, decision and type labeling, the Monte Carlo simulator continues to perform the next sampling and iterates until a total of 10,000 samplings are completed. Then, the current round of simulation sampling and decision-making is stopped, and the key values ​​in the simulation process are then counted. First, the total number of all behaviors marked as shutdown triggers is counted. This value is the sum of the number of valid triggers and the number of false triggers. Then, the number of all behaviors marked as false triggers is counted. This value only includes the number of times normal fluctuation samples are incorrectly judged as shutdown triggers.

[0110] (6) Calculate the false trigger probability. The calculation rule is to divide the number of false triggers by the total number of shutdown triggers. The resulting value is the false trigger probability corresponding to the current threshold. The calculation result is directly used for comparison and judgment with the preset probability threshold.

[0111] S3.4: Compare the current false trigger probability with the preset probability threshold of 1%; if the current false trigger probability is greater than the preset probability threshold, start the particle swarm optimization algorithm: take the first-level warning threshold, the second-level load reduction threshold, and the third-level emergency shutdown threshold as variables to be optimized, take the current false trigger probability as the fitness function, take minimizing the false trigger probability as the optimization objective, and search to obtain the optimal set of graded shutdown protection thresholds; replace the current threshold with the optimal set of graded shutdown protection thresholds, and iterate repeatedly until the current false trigger probability is lower than the preset probability threshold, then stop the iteration;

[0112] Furthermore, the specific steps in S3.4 include:

[0113] (1) Compare the current false trigger probability calculated by Monte Carlo simulation with the preset probability threshold set by the system. The preset probability threshold is fixed at 1%, which is the highest false trigger level allowed by the system and is the only standard for determining whether to start the particle swarm optimization algorithm. When comparing, directly read the calculation result of the current false trigger probability. When the value of the current false trigger probability is greater than 1%, it is determined that the current threshold cannot meet the protection accuracy requirements and the particle swarm optimization algorithm needs to be started to optimize the threshold. When the value of the current false trigger probability is less than or equal to 1%, it is determined that the current threshold meets the usage requirements and no optimization operation is required.

[0114] (2) After confirming the need to start the particle swarm optimization algorithm, the parameters are configured first. The population size is set to 30, meaning that there are 30 independent threshold combinations participating in the optimization process in each iteration. Each threshold combination includes three variables to be optimized: a first-level warning threshold, a second-level load reduction threshold, and a third-level emergency shutdown threshold. The maximum number of iterations is set to 100, which ensures that the algorithm converges fully and avoids wasting computational resources due to invalid iterations. The inertia weight is set to 0.7 to control the influence of the particle's historical velocity on the current velocity and to balance the algorithm's global search capability and local search capability. The individual learning factor is set to... The social learning factor is set to 1.5, and the two learning factors are equal to ensure that the particle takes into account both its own optimal position and the population optimal position during the optimization process; the upper limit of the particle's speed is set to 5% of the original value of the corresponding threshold, and the lower limit of the speed is set to -5% of the original value of the corresponding threshold to avoid numerical mutations during particle updates that could cause threshold logic confusion; the upper and lower limits of the particle's position are preset according to the actual operating parameter range of the equipment to ensure that the optimized threshold is always within a reasonable and feasible range, while satisfying the incremental logic constraint that the first-level warning threshold is less than the second-level load reduction threshold, and the second-level load reduction threshold is less than the third-level emergency shutdown threshold.

[0115] (3) When the algorithm is initialized, the set of the initially adjusted hierarchical shutdown protection thresholds is used as the initial position of all particles. Each particle corresponds to an independent set of thresholds. The initial velocity of all particles is set to zero. Then, the fitness value corresponding to each particle is calculated. The fitness function is directly adopted using the false trigger probability obtained by Monte Carlo simulation. The optimization objective is to minimize the fitness value. That is, the lower the false trigger probability, the better the fitness of the particle. In each iteration, Monte Carlo simulation is first performed on the threshold combination corresponding to the current position of each particle to calculate the corresponding false trigger probability as the current fitness value. The current fitness value is compared with the particle's own historical best fitness value. If the current fitness value is better, the particle's own historical best position and best fitness value are updated. At the same time, the current fitness values ​​of all particles are compared globally to find the particle with the best fitness in the current population and update its position and fitness value to the global best position and global best fitness value.

[0116] (4) After the optimal value update is completed, the flight speed of each particle is updated according to the preset inertia weight, individual learning factor and social learning factor. Then, the position of each particle is adjusted according to the updated flight speed. Immediately after the position is updated, incremental logic verification is performed to ensure that each set of threshold combinations maintains the logical relationship that the first-level warning threshold is less than the second-level load reduction threshold and the second-level load reduction threshold is less than the third-level emergency stop threshold. The position of the particle that fails the logic verification will be forcibly pulled back to the nearest legal position. After a single iteration is completed, it is determined whether the maximum number of iterations has been reached or whether a threshold combination with a false trigger probability of less than 1% has been found. If either condition is met, the current algorithm optimization is stopped.

[0117] (5) After the particle swarm optimization algorithm stops, the threshold combination corresponding to the global optimal position is output as the optimal hierarchical shutdown protection threshold set. Then, the current threshold of the previous round is completely replaced by the optimal hierarchical shutdown protection threshold set. The new current threshold is input into the Monte Carlo simulator again, and the corresponding current false trigger probability is re-sampled and calculated. The newly obtained current false trigger probability is compared with the 1% preset probability threshold again. If the current false trigger probability is still greater than the preset probability threshold, the particle swarm optimization algorithm is restarted and iterated repeatedly to search for a better hierarchical shutdown protection threshold set. If the current false trigger probability is lower than the preset probability threshold, the threshold optimization is determined to be completed, and all iteration processes are stopped immediately. The threshold combination at this time is the final hierarchical shutdown protection threshold set that meets the protection accuracy requirements.

[0118] S3.5: When the iteration stops, the final optimal set of hierarchical shutdown protection thresholds is written into the non-volatile memory, and incremental logic is performed on the first-level warning threshold, the second-level load reduction threshold, and the third-level emergency shutdown threshold to ensure that the first-level warning threshold is less than the second-level load reduction threshold and the second-level load reduction threshold is less than the third-level emergency shutdown threshold. After the incremental logic verification passes, a protection strategy ready signal containing the system time and threshold version number is generated.

[0119] It should be noted that the Level 1 warning threshold corresponds to minor anomalies and should be the lowest; the Level 2 load reduction threshold corresponds to moderate anomalies and should be higher than the Level 1 warning threshold; the Level 3 emergency shutdown threshold corresponds to severe anomalies and should be the highest. If the verification fails, the system will automatically restart the optimization iteration until the threshold logic is correct.

[0120] The specific steps of S4 include:

[0121] S4.1: Based on the protection strategy ready signal, monitor the oil status parameters and operating condition parameters in real time. When any monitored parameter reaches the level shutdown condition corresponding to the optimal level shutdown protection threshold, execute the corresponding shutdown action according to the determined level. Among them, triggering the first level early warning only outputs an alarm signal, triggering the second level load reduction reduces the motor output power, and triggering the third level emergency shutdown sends a hard wire cut-off signal to the motor drive circuit.

[0122] Furthermore, the specific steps of S4.1 include:

[0123] (1) After receiving the protection strategy ready signal after verification is completed and generated, the system immediately enters the real-time protection execution state and starts the uninterrupted real-time monitoring process of oil status parameters and operating condition parameters. The real-time monitoring refresh cycle is fixed at 50ms. This cycle can ensure that parameter changes are captured in time and will not bring too high computational load to the system control unit. The monitoring objects include all oil status parameters and operating condition parameters. Among them, the oil status parameters include oil dielectric constant, oil viscosity index, oil water content and oil particle concentration. The operating condition parameters include component outlet pressure, component vibration acceleration, effective value of motor stator current and main shaft speed of conveying component. All parameters are effective values ​​after filtering and time synchronization processing during monitoring.

[0124] (2) During the monitoring process, the hierarchical shutdown protection threshold after iterative optimization and logical verification will be continuously called. The hierarchical shutdown protection threshold includes three items: the first-level early warning threshold, the second-level load reduction threshold, and the third-level emergency shutdown threshold. The three thresholds maintain a fixed increasing relationship and will not be modified during the real-time execution phase. The system will compare the value of each parameter obtained from real-time monitoring with the hierarchical threshold of the corresponding parameter to determine whether the shutdown trigger condition of the corresponding level is met. The shutdown trigger condition adopts the single parameter over-limit judgment rule. That is, when the real-time monitoring value of any parameter exceeds the upper limit of the threshold of the corresponding level or is lower than the lower limit of the threshold of the corresponding level, the shutdown condition of that level is immediately determined to be met.

[0125] (3) When the system determines that the monitoring parameter only reaches the triggering condition corresponding to the first-level warning threshold and does not reach the triggering condition of a higher level, the system executes the protection action corresponding to the first-level warning. The first-level warning action is only a status prompt and information report, and does not interfere with the normal operation of the equipment. The system drives the sound and light alarm device installed on site to emit a low-frequency sound and light prompt. The light prompt color is set to yellow. At the same time, the warning information is uploaded to the remote monitoring platform through the industrial communication bus. The warning information includes the current time, monitoring point number, parameter name that triggers the warning, real-time parameter value and first-level warning threshold value. The entire first-level warning process only outputs the alarm signal, does not change the motor operating status, and does not adjust the working parameters of the material conveying component.

[0126] (4) When the system determines that the monitoring parameters reach the triggering conditions corresponding to the secondary load reduction threshold, regardless of whether the primary warning is triggered, the protection action corresponding to the secondary load reduction is executed first. The secondary load reduction action aims to reduce the equipment load and slow down the development of the abnormality. The system sends a load reduction control command to the frequency converter matched with the motor. The load reduction range is fixed at 40% of the rated output power. After receiving the load reduction control command, the frequency converter gradually reduces the output power of the motor within a smooth transition time of 3 seconds. At the same time, it simultaneously reduces the spindle speed of the material conveying component to avoid mechanical shock caused by the sudden drop in power. After the load reduction is completed, the motor continues to run at 60% of the rated power. The system maintains the load reduction state and continuously reports the load reduction operation information, waiting for manual intervention or automatic parameter recovery.

[0127] (5) When the system determines that the monitoring parameters reach the triggering conditions corresponding to the level 3 emergency stop threshold, the system immediately ignores the action logic of the first two levels and directly executes the protection action corresponding to the level 3 emergency stop. The level 3 emergency stop is the highest priority safety protection action. The system sends a hard wire cut-off signal to the hardware actuator of the motor drive circuit. The hard wire cut-off signal is a hard wire level signal independent of the communication bus, which can directly physically disconnect the power supply circuit and control circuit of the motor. The response time of the hard wire cut-off signal is set to within 10ms to ensure that the motor stops running in the shortest time and the material conveying component stops working completely to avoid the abnormal state from continuing to expand and causing equipment damage or safety accidents. After the emergency stop action is executed, the system maintains the hardware cut-off state until the fault is cleared and a manual reset command is received.

[0128] (6) Throughout the entire graded action execution process, the system strictly follows the action priority logic. The third-level emergency shutdown has the highest priority, followed by the second-level load reduction, and the first-level early warning has the lowest priority. When multiple grade conditions are met at the same time, the system only executes the action corresponding to the highest priority to ensure that the protection action is accurate and does not conflict. Each action execution will generate an independent event record. The record content includes the action trigger time, trigger level, trigger parameters, real-time parameter values ​​and corresponding threshold values. The event record will be synchronously stored in the local control unit and the remote monitoring platform.

[0129] S4.2: While performing a level 3 emergency shutdown and sending a hard-wire cut-off signal, start a power supply maintenance window with a preset delay duration, and divide the power supply maintenance window into an initial period and a middle period;

[0130] Furthermore, at the same moment that the Level 3 emergency stop action is executed and a hard-wire cut-off signal is sent out, the start command for the power supply maintenance window is triggered simultaneously. This ensures that the stop action and the power supply maintenance operation are completely synchronized in time, with no timing deviation. The preset delay duration of the power supply maintenance window is fixed in the system before it leaves the factory, with an overall duration of 3 seconds. This duration can fully cover the entire transient process from equipment operation to shutdown, without causing ineffective consumption of power resources. After the power supply maintenance window is started, the system automatically cuts off the power supply to the motor drive circuit and non-essential auxiliary components, continuing to supply power only to the oil state parameters and operating conditions. All sensors corresponding to the condition parameters are provided with a stable power supply. The sensor power supply voltage is maintained at the rated DC 24V to ensure that all types of sensors can still collect signals normally during the shutdown transient and there will be no data loss due to the power failure of the whole machine. In this embodiment, the initial period is the first 1 second after the window opens. It is the stage where the parameters fluctuate violently and the abnormal transient characteristics are most prominent at the moment of shutdown. It is used to capture the extreme value data of the fault. The middle period is the middle interval from the end of the initial period to the end of the window countdown. At this time, the equipment vibration and impact have initially decayed and the oil parameters tend to be stable, which is suitable for collecting steady-state average values ​​to characterize the stable oil circuit conditions after shutdown.

[0131] S4.3: During the power supply window, the power supply to each sensor corresponding to the oil state parameters and operating condition parameters is maintained continuously. During the initial period of the power supply window, the oil particle concentration and component vibration acceleration are collected at a preset sampling interval. The maximum values ​​are taken as the peak value of the oil particle concentration and the peak value of the vibration acceleration during shutdown. At the same time, the oil dielectric constant is collected and the maximum value is taken as the peak value of the dielectric constant during shutdown.

[0132] Furthermore, the specific steps of S4.3 include:

[0133] (1) The initial period of the power supply window is fixed in the configuration as the 0s to 1s interval after the window is opened. This period is the stage with the most drastic parameter changes and the most obvious abnormal characteristics during the equipment shutdown process. It is used to collect transient peak data. During this period, the system starts the high-speed acquisition process according to the preset sampling interval. The sampling interval is set to 1ms, that is, data sampling is completed once every 1ms. 1000 effective samples can be completed continuously during the entire initial period to ensure that the instantaneous extreme values ​​of the parameters can be captured completely. The system continuously collects two parameters: oil particle concentration and component vibration acceleration. Each value is stored in the temporary buffer area in real time. After the initial period ends, all 1000 sets of sampled values ​​stored in the buffer area are compared one by one. The largest value is selected and the maximum value is used as the peak value of the transient oil particle concentration and the peak value of the transient vibration acceleration during shutdown.

[0134] (2) During the same time period of collecting oil particle concentration and component vibration acceleration, the same rules are used to collect the oil dielectric constant. 1000 samples are collected in the initial period at a sampling interval of 1ms. The sampled data is stored in an independent temporary buffer in real time. After the initial period ends, all sampled values ​​in the buffer are compared one by one, and the one with the largest value is extracted as the peak value of the transient dielectric constant during shutdown. All transient peak data are temporarily locked after extraction and are not allowed to be overwritten by subsequent data to ensure the uniqueness and accuracy of the data. All transient peak parameters are the most extreme state representation at the moment of shutdown and can truly reflect the severity of the abnormality.

[0135] S4.4: During the middle period of the power supply window, the oil viscosity index and oil moisture content are continuously sampled and the arithmetic mean is calculated. These are used as the steady-state values ​​of the shutdown viscosity index and the shutdown moisture content, respectively. The formula for calculating the arithmetic mean is the prior art in this field and is not an inventive solution of this application. It will not be elaborated here. In this embodiment, the middle period of the power supply window is the interval between the end of the initial period and the end of the window countdown.

[0136] S4.5: Generate a unique shutdown event number for this shutdown action, and encapsulate the peak values ​​of transient oil particle concentration, transient vibration acceleration, and transient dielectric constant, as well as the steady-state values ​​of viscosity index and moisture content, in a structured manner, and output a latch-up judgment data packet.

[0137] The specific steps of S5 include:

[0138] S5.1: Extract the steady-state value of shutdown moisture content, the peak value of shutdown transient oil particle concentration, the peak value of shutdown transient dielectric constant, and the steady-state value of shutdown viscosity index from the lockout judgment data packet. Compare them with the preset upper and lower thresholds of moisture content, particle concentration, dielectric constant, and viscosity index in the qualified oil state range. When the steady-state value of shutdown moisture content is lower than the upper threshold of moisture content, the peak value of shutdown transient oil particle concentration is lower than the upper threshold of particle concentration, the peak value of shutdown transient dielectric constant is between the upper and lower thresholds of dielectric constant, and the steady-state value of shutdown viscosity index is between the upper and lower thresholds of viscosity index, output the qualified oil state signal. Otherwise, output the judgment result that the oil state has not been restored and maintain the shutdown lockout state.

[0139] It should be noted that the acceptable range for oil condition is a range set according to oil product standards, equipment lubrication requirements, and safe operation specifications. Only when all core oil indicators fall within the acceptable range can the basic performance of the oil be determined to meet the restart requirements. The steady-state value of the shutdown moisture content reflects the dryness of the oil; if it is below the upper threshold, it indicates that moisture has been eliminated. The peak value of the transient oil particle concentration at shutdown reflects the cleanliness of the oil; if it is below the upper threshold, it indicates that particulate impurities have been removed. If the peak value of the transient dielectric constant at shutdown and the steady-state value of the shutdown viscosity index are within the acceptable range, it indicates that the overall performance and lubrication performance of the oil have returned to normal. If all four indicators meet the conditions, the system outputs an acceptable signal for the basic oil condition. If any one indicator is not met, it is determined that the oil condition has not been restored, the system remains locked, and restart is prohibited.

[0140] Furthermore, the specific steps in S5.1 include:

[0141] (1) Read and parse the lockout judgment data packet that has been structured and encapsulated. Extract four core parameters for oil basic state determination from the lockout judgment data packet in the order of data reading. These parameters are the steady-state value of shutdown moisture content, the peak value of shutdown transient oil particle concentration, the peak value of shutdown transient dielectric constant, and the steady-state value of shutdown viscosity index. Only read the values ​​during the extraction process.

[0142] (2) After the parameter extraction is completed, the oil condition qualified range parameters stored in the non-volatile memory are called. The oil condition qualified range is fixed according to the equipment lubrication standard and oil usage specifications. All thresholds are not dynamically adjusted during operation. Among them, the upper limit threshold of the water content qualified range is set to 0.05%, the upper limit threshold of the particle concentration qualified range is set to 50,000 particles / L, the lower limit threshold of the dielectric constant qualified range is set to 2.0, the upper limit threshold is set to 3.5, and the lower limit threshold of the viscosity index qualified range is set to 90, the upper limit threshold is set to 110.

[0143] (3) Start performing item-by-item numerical comparison and judgment. First, judge the steady-state value of the shutdown moisture content. Compare the extracted steady-state value of the shutdown moisture content with the preset upper limit threshold of moisture content. Only when the value of the steady-state value of the shutdown moisture content is less than the upper limit threshold of moisture content, the judgment result of this item is qualified. If the steady-state value of the shutdown moisture content is greater than or equal to the upper limit threshold of moisture content, the judgment result of this item is unqualified.

[0144] (4) Determine the peak value of the transient oil particle concentration during shutdown. Compare the extracted peak value of the transient oil particle concentration during shutdown with the preset upper limit threshold of particle concentration. Only when the value of the peak value of the transient oil particle concentration during shutdown is less than the upper limit threshold of particle concentration, the result of this determination is qualified. If the peak value of the transient oil particle concentration during shutdown is greater than or equal to the upper limit threshold of particle concentration, the result of this determination is unqualified.

[0145] (5) Determine the peak value of the transient dielectric constant during shutdown. Compare the extracted peak value of the transient dielectric constant with the preset lower and upper threshold values ​​of the dielectric constant. Only when the value of the peak value of the transient dielectric constant during shutdown is greater than the lower threshold value and less than the upper threshold value of the dielectric constant, the determination result is qualified. If the peak value of the transient dielectric constant during shutdown is less than or equal to the lower threshold value or greater than or equal to the upper threshold value, the determination result is unqualified.

[0146] (6) Determine the steady-state value of the viscosity index at shutdown. Compare the extracted steady-state value of the viscosity index at shutdown with the preset lower threshold and upper threshold of the viscosity index. Only when the value of the steady-state value of the viscosity index at shutdown is greater than the lower threshold of the viscosity index and less than the upper threshold of the viscosity index, the result of this determination is qualified. If the steady-state value of the viscosity index at shutdown is less than or equal to the lower threshold or greater than or equal to the upper threshold, the result of this determination is unqualified.

[0147] (7) The judgment of the four parameters is based on the logical AND rule. The system will determine that the basic state of the oil meets the recovery requirements and output the basic state of the oil is qualified only if the individual judgment results of the four parameters are qualified at the same time. If any one of the four parameters is unqualified, the system will determine that the oil state has not been restored regardless of whether the other parameters are qualified. The system will directly output the judgment result that the oil state has not been restored, and at the same time, the current shutdown lockout state will remain unchanged. The equipment will not be allowed to restart, nor will the hardware lockout protection be released.

[0148] S5.2: When the output oil basic state qualified signal, extract the complete component vibration acceleration time sequence sampling sequence that is from the same source as the peak value of the transient vibration acceleration during shutdown in the lockout judgment data packet, perform fast Fourier transform on the component vibration acceleration time sequence sampling sequence to obtain the frequency spectrum, and calculate the ratio of the sum of the energy amplitudes corresponding to the preset gear meshing frequency and its harmonics in the frequency spectrum to the total energy amplitude of the entire frequency band to obtain the main frequency energy ratio. When the main frequency energy ratio is lower than the preset background noise energy ratio threshold, output the mechanical shock attenuated signal; otherwise, output the judgment result that the oil state has not recovered and maintain the shutdown lockout state.

[0149] Furthermore, the specific steps in S5.2 include:

[0150] (1) Extract the complete component vibration acceleration time-series sampling sequence from the locking determination data packet. The component vibration acceleration time-series sampling sequence is the vibration signal continuously collected during the shutdown process. The data length is fixed at 1024 sampling points and the sampling frequency is set to 10kHz, which can completely cover the entire process of vibration decay.

[0151] (2) Perform Fast Fourier Transform on the extracted component vibration acceleration time-series sampling sequence. Before the transformation, the time-series sampling sequence is windowed. The window function used is the Hanning window. The Hanning window can effectively reduce spectral leakage and improve the accuracy of frequency analysis. After the windowing is completed, the Fast Fourier Transform calculation is started to convert the vibration signal that was originally in the time domain into a frequency domain representation to obtain complete frequency spectrum data. The frequency spectrum contains the energy amplitude corresponding to all frequency components from low frequency to high frequency, which can clearly reflect the frequency distribution characteristics of the vibration signal. The Fast Fourier Transform and the Hanning window are existing technologies in this field and are not the inventive solutions of this application. They will not be described in detail here.

[0152] (3) After completing the frequency spectrum calculation, the pre-stored gear meshing frequency parameters are called. The gear meshing frequency parameters are calibrated before leaving the factory according to the spindle speed of the conveying component and the number of gear teeth. The gear meshing frequency is fixed at 200Hz, and its multiples are set to 400Hz, 600Hz and 800Hz respectively. These frequencies are the inherent characteristic frequencies of the gears inside the conveying component when they are running, and are also the core basis for judging whether mechanical impact exists. The system accurately locates the positions of the above gear meshing frequencies and their multiples in the frequency spectrum, extracts the energy amplitude at these frequency points one by one, and then accumulates all the extracted energy amplitudes to obtain the sum of the main frequency characteristic energy amplitudes.

[0153] (4) At the same time, the energy amplitude of all frequency bands in the frequency spectrum is accumulated to obtain the total energy amplitude of the whole frequency band. This value includes the energy of all frequency components in the vibration signal, including mechanical characteristic frequency energy, as well as environmental noise and random interference energy. Then, the proportion of the main frequency energy is calculated. The calculation method is to divide the sum of the main frequency characteristic energy amplitudes by the total energy amplitude of the whole frequency band. The ratio obtained is the proportion of the main frequency energy of the current vibration signal. This ratio can intuitively reflect the proportion of mechanical characteristic frequency energy in the overall vibration energy.

[0154] (5) Call the preset background noise energy ratio threshold. The background noise energy ratio threshold is set to 10%. Compare the calculated main frequency energy ratio with the background noise energy ratio threshold. When the value of the main frequency energy ratio is less than 10%, it means that the energy amplitude corresponding to the gear meshing frequency and its harmonics is extremely low. The mechanical shock and abnormal vibration have been completely attenuated. The equipment has returned to a stable state without shock. The system immediately outputs a signal that the mechanical shock has been attenuated. When the value of the main frequency energy ratio is greater than or equal to 10%, it means that there is still obvious gear shock or structural vibration inside the machine. The shock state has not been completely eliminated. The system directly outputs the judgment result that the oil state has not been restored. At the same time, the shutdown lockout state is maintained. The protection is not released and the equipment is not allowed to be restarted.

[0155] S5.3: When the output oil basic condition is qualified and the mechanical shock has decayed, the output oil condition has been restored. The shutdown lockout state is released and the system returns to S4 to restart real-time monitoring. Otherwise, the shutdown lockout state is maintained.

[0156] Following the determination that the output oil state has not recovered, a retest based on periodic low-speed idling is also included, specifically:

[0157] A1: After outputting the judgment result that the oil condition has not been restored and maintaining the shutdown lockout state, start the periodic retest timer. The preset duration of the periodic retest timer is the waiting period for the retest.

[0158] Furthermore, the preset duration of the periodic re-inspection timer is set according to the difficulty of oil recovery and the type of fault, usually ranging from 10 minutes to 30 minutes. This avoids the extra burden on the oil circuit caused by frequent testing and can detect the recovery status of the oil in a timely manner. After the periodic re-inspection timer is started, the system remains shut down and locked, only maintaining power supply to the sensors and control units, waiting for the re-inspection opportunity.

[0159] A2: When the periodic re-inspection timer expires, the pressure relief bypass valve pre-installed at the oil inlet of the conveying component is opened to form a low-pressure circulation loop in the oil circuit. The motor is driven by the frequency converter to run at 5% to 10% of the rated speed for 5 to 10 minutes. During this period, the oil particle concentration, oil moisture content and component outlet pressure are collected again to form re-inspection lockout judgment data. In this embodiment, during the idling process, the system collects the oil particle concentration, oil moisture content and component outlet pressure in real time. The oil particle concentration reflects the cleanliness recovery, the oil moisture content reflects the dryness recovery, and the component outlet pressure reflects whether the oil circuit is full and leak-free. The three parameters form the re-inspection lockout judgment data for judging the status of this re-inspection.

[0160] A3: Compare the oil particle concentration in the output re-inspection lockout judgment data with the upper limit threshold of particle concentration in the qualified oil condition range, compare the oil moisture content in the re-inspection lockout judgment data with the upper limit threshold of moisture content in the qualified oil condition range, and compare the component outlet pressure in the re-inspection lockout judgment data with the preset minimum start-up outlet pressure threshold.

[0161] Furthermore, the specific steps in A3 include:

[0162] (1) After completing the periodic low-speed idling re-inspection test, three key judgment parameters are extracted from the re-inspection lockout judgment data generated in this test in a fixed order. These parameters are oil particle concentration value, oil moisture content value and component outlet pressure value. All three parameters are valid values ​​collected in real time during the idling test and processed by steady-state filtering.

[0163] (2) After the parameter extraction is completed, the system retrieves the pre-set oil condition qualified range threshold and equipment start-up pressure threshold from the non-volatile memory. The upper limit threshold of particle concentration in the oil condition qualified range is set to 50,000 particles / L, the upper limit threshold of moisture content in the oil condition qualified range is set to 0.05%, and the preset minimum start-up outlet pressure threshold is set to 0.2MPa.

[0164] (3) The oil particle concentration value extracted from the re-inspection lockout judgment data is compared one-to-one with the upper limit threshold of particle concentration in the qualified oil condition range, and only the size of a single dimension is judged; the oil moisture content value extracted from the re-inspection lockout judgment data is compared one-to-one with the upper limit threshold of moisture content in the qualified oil condition range; the component outlet pressure value extracted from the re-inspection lockout judgment data is compared one-to-one with the preset minimum start-up outlet pressure threshold; after the comparison is completed, the system temporarily stores the independent comparison result of each parameter and does not perform comprehensive logical calculation.

[0165] A4: When the oil particle concentration is lower than the upper limit threshold for particle concentration, the oil moisture content is lower than the upper limit threshold for moisture content, and the component outlet pressure is higher than the minimum start-up outlet pressure threshold, the output will indicate that the oil condition has been restored after the re-inspection. This indicates that after this round of low-speed idling cycle, the cleanliness, dryness, and oil supply status of the oil have all reached the standard for re-entering the mechanical condition judgment. Otherwise, the cycle re-inspection timer will be reset and the system will return to A1 to wait for the next re-inspection test.

[0166] Furthermore, when the overall judgment result is unqualified, the system does not output any recovery signal, but directly executes the reset operation of the periodic re-inspection timer. The preset waiting time of the periodic re-inspection timer is fixed at 15 minutes. The reset operation will clear the timing data of the periodic re-inspection timer and restart the countdown. After the periodic re-inspection timer is reset, the system automatically returns to the initial waiting state and continues to maintain the shutdown and lockout state, waiting for the start time of the next re-inspection test. During the waiting period, the system only maintains the basic power supply to the sensor, does not start the motor, does not open the pressure relief bypass valve, and does not perform any oil circuit circulation actions, until the periodic re-inspection timer reaches the set waiting time again, and then restarts the next round of low-speed idling re-inspection test, until all three parameters meet the qualified conditions and output the judgment result that the re-inspection oil state has been restored, ensuring that the equipment is always in a safe lockout state when the oil state is not fully up to standard, and there will be no situation of accidental start or operation with faults.

[0167] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

Claims

1. A method for abnormal shutdown protection of a material conveying assembly based on oil state sensing, characterized in that, include: S1: Collect the oil state parameters in the working circuit of the material conveying component and the operating condition parameters of the material conveying component; S2: Based on the oil state parameters and the operating condition parameters, construct a dual-channel deep confidence network model and output abnormal feature judgment data; S3: Based on the abnormal feature determination data, perform adaptive iterative correction of the graded shutdown protection threshold until the false trigger probability of triggering shutdown action based on the obtained optimal graded shutdown protection threshold is lower than the preset probability threshold, and output the protection strategy ready signal. S4: Based on the protection strategy ready signal, monitor the oil status parameters and operating condition parameters in real time. When the monitored oil status parameters and operating condition parameters reach the level shutdown condition corresponding to the optimal level shutdown protection threshold, trigger the corresponding shutdown action and collect the lockout judgment data after shutdown. S5: Based on the lockout determination data, determine whether the oil condition has recovered to the preset oil condition qualified range. If it has not recovered, maintain the shutdown lockout state; if it has recovered, release the lockout and restart monitoring. The specific steps of S3 include: Read the pre-stored set of graded shutdown protection thresholds; the set of graded shutdown protection thresholds includes a first-level early warning threshold, a second-level load reduction threshold, and a third-level emergency shutdown threshold; The system analyzes the severity level of anomalies in the data and assigns initial adjustment coefficients to the first-level warning threshold, the second-level load reduction threshold, and the third-level emergency shutdown threshold according to the pre-stored level threshold mapping table. The system then outputs the set of initially adjusted level shutdown protection thresholds. The initial adjusted set of hierarchical shutdown protection thresholds is used as the current threshold. The historical abnormal feature judgment data sequence obtained by cumulative storage is used as the sample and input into the Monte Carlo simulator. Each time, a sample is randomly selected and a shutdown trigger decision is made according to the current threshold. The total number of shutdowns triggered and the number of false triggers are counted to calculate the current false trigger probability. The current false trigger probability is compared with a preset probability threshold. If the current false trigger probability is greater than the preset probability threshold, the particle swarm optimization algorithm is started: the first-level warning threshold, the second-level load reduction threshold, and the third-level emergency shutdown threshold are used as variables to be optimized, and the current false trigger probability is used as the fitness function. The optimal set of graded shutdown protection thresholds is searched to minimize the false trigger probability. The current threshold is replaced with the optimal set of graded shutdown protection thresholds, and the iteration is repeated until the current false trigger probability is lower than the preset probability threshold. When the iteration stops, the final optimal set of hierarchical shutdown protection thresholds is written into non-volatile memory, and incremental logic is performed on the first-level warning threshold, the second-level load reduction threshold, and the third-level emergency shutdown threshold to ensure that the first-level warning threshold is less than the second-level load reduction threshold and the second-level load reduction threshold is less than the third-level emergency shutdown threshold. After the incremental logic verification passes, a protection strategy ready signal containing the system time and threshold version number is generated.

2. The method for abnormal shutdown protection of a material conveying assembly based on oil state sensing as described in claim 1, characterized in that, The specific steps of S2 include: Read the oil state parameters and the operating condition parameters of the conveying assembly; the oil state parameters include at least the oil dielectric constant, oil viscosity index, oil moisture content and oil particle concentration; the operating condition parameters include at least the assembly outlet pressure, assembly vibration acceleration, motor stator current RMS value and conveying assembly spindle speed; For the time-series data of oil particle concentration in the oil state parameters, a three-level decomposition is performed using Daubechies fourth-order wavelet packet decomposition to extract the energy proportion of each frequency band node, forming a frequency domain energy feature vector of particle concentration; for the time-series data of oil viscosity index in the oil state parameters, the sliding window average value is calculated and the first-order backward difference is obtained to obtain the viscosity change trend feature; the oil dielectric constant, oil moisture content, particle concentration frequency domain energy feature vector, and viscosity change trend feature are horizontally concatenated to output the oil degradation feature vector. A short-time Fourier transform is performed on the component vibration acceleration in the operating condition parameters to obtain the vibration time-frequency feature matrix. The component outlet pressure, effective value of motor stator current, and spindle speed of the material conveying component in the operating condition parameters are normalized and then superimposed on the vibration time-frequency feature matrix in the channel dimension to output the operating condition feature matrix.

3. The method for abnormal shutdown protection of a material conveying assembly based on oil state sensing as described in claim 2, characterized in that, The specific steps of S2 also include: A dual-channel deep belief network model is constructed. The first channel is a stacked structure of three restricted Boltzmann machines, which takes the oil degradation feature vector as input and outputs the first deep feature vector. The second channel is a time-frequency conversion network consisting of two convolutional layers followed by two pooling layers, which takes the operating condition feature matrix as input and outputs the second deep feature vector. The first deep feature vector and the second deep feature vector are concatenated in a feature concatenation layer to output a joint feature vector. The joint feature vector is sequentially input into two fully connected layers and the output layer to calculate the health index value at the current time. The health index value is compared with the preset health baseline threshold. When the health index value is lower than the health baseline threshold and the difference exceeds the preset deviation threshold, abnormal feature judgment data is output. The abnormal feature judgment data includes an abnormality type label and an abnormality severity level.

4. The method for abnormal shutdown protection of a material conveying assembly based on oil state sensing as described in claim 1, characterized in that, The specific steps of S4 include: Based on the protection strategy ready signal, the oil status parameters and operating condition parameters are monitored in real time. When any of the monitored parameters reaches the level shutdown condition corresponding to the optimal level shutdown protection threshold, the corresponding shutdown action is executed according to the determined level. Among them, triggering the first level early warning only outputs an alarm signal, triggering the second level reduces the motor output power, and triggering the third level emergency shutdown sends a hard wire cut-off signal to the motor drive circuit. While executing a level 3 emergency shutdown and sending a hard-wire cut-off signal, a power supply maintenance window with a preset delay duration is initiated, and the power supply maintenance window is divided into an initial period and a middle period.

5. The method for abnormal shutdown protection of a material conveying assembly based on oil state sensing as described in claim 4, characterized in that, The specific steps of S4 also include: Within the power supply window, power is continuously supplied to the sensors corresponding to the oil state parameters and operating condition parameters. During the initial period of the power supply window, the oil particle concentration and component vibration acceleration are collected at a preset sampling interval. The maximum values ​​are taken as the peak values ​​of the transient oil particle concentration and the transient vibration acceleration during shutdown. At the same time, the oil dielectric constant is collected and the maximum value is taken as the peak value of the transient dielectric constant during shutdown. During the middle period of the power supply window, the oil viscosity index and oil moisture content were continuously sampled and the arithmetic mean was calculated, which were used as the steady-state values ​​of the shutdown viscosity index and the shutdown moisture content, respectively. A unique shutdown event number is generated for this shutdown action. The peak values ​​of transient oil particle concentration, transient vibration acceleration, and transient dielectric constant during shutdown, as well as the steady-state values ​​of viscosity index and moisture content during shutdown, are structured and encapsulated, and a latch-up judgment data packet is output.

6. The method for abnormal shutdown protection of a material conveying assembly based on oil state sensing as described in claim 1, characterized in that, The specific steps of S5 include: The steady-state value of shutdown moisture content, the peak value of shutdown transient oil particle concentration, the peak value of shutdown transient dielectric constant, and the steady-state value of shutdown viscosity index are extracted from the lockout judgment data packet. They are compared with the preset upper and lower thresholds of moisture content, particle concentration, dielectric constant, and viscosity index in the qualified oil state range. When the steady-state value of shutdown moisture content is lower than the upper threshold of moisture content, the peak value of shutdown transient oil particle concentration is lower than the upper threshold of particle concentration, the peak value of shutdown transient dielectric constant is between the upper and lower thresholds of dielectric constant, and the steady-state value of shutdown viscosity index is between the upper and lower thresholds of viscosity index, the basic state of oil is qualified. Otherwise, the judgment result that the oil state has not been restored is output and the shutdown lockout state is maintained. When the output signal indicating that the basic state of the oil is qualified, the component vibration acceleration time-series sampling sequence that is from the same source as the peak value of the transient vibration acceleration during shutdown is extracted from the lockout judgment data packet. The component vibration acceleration time-series sampling sequence is subjected to fast Fourier transform to obtain the frequency spectrum. The ratio of the sum of the energy amplitudes corresponding to the preset gear meshing frequency and its harmonics in the frequency spectrum to the total energy amplitude of the entire frequency band is calculated to obtain the main frequency energy ratio. When the main frequency energy ratio is lower than the preset background noise energy ratio threshold, the mechanical shock attenuation signal is output. Otherwise, the judgment result that the oil state has not been restored is output and the shutdown lockout state is maintained. When the output oil condition is qualified and the mechanical shock has decayed, the output oil condition has been restored. The shutdown lockout state is released and the system returns to S4 to restart real-time monitoring. Otherwise, the shutdown lockout state is maintained.

7. The method for abnormal shutdown protection of a material conveying assembly based on oil state sensing as described in claim 6, characterized in that, Following the determination that the output oil state has not recovered, a retest based on periodic idling is also included, specifically: After outputting the judgment result that the oil condition has not been restored and maintaining the shutdown lockout state, the periodic re-inspection timer is started. The preset duration of the periodic re-inspection timer is the waiting period for the re-inspection test. When the periodic re-inspection timer expires, the pressure relief bypass valve pre-installed at the oil inlet of the conveying component is opened, so that the oil circuit forms a low-pressure circulation loop. The motor is driven by the frequency converter to run at a preset percentage of the rated speed as the idling speed for a preset cleaning time, and the oil particle concentration, oil moisture content and component outlet pressure are collected again to form re-inspection lockout judgment data. The oil particle concentration, oil moisture content and component outlet pressure are used to judge the oil cleanliness, dryness and oil circuit fullness, respectively.

8. The method for abnormal shutdown protection of a material conveying assembly based on oil state sensing as described in claim 7, characterized in that, Following the determination that the output oil state has not recovered, a retest based on periodic idling is also included, specifically including: The oil particle concentration in the output re-inspection lockout judgment data is compared with the upper limit threshold of particle concentration in the qualified oil condition range. The water content in the re-inspection lockout judgment data is compared with the upper limit threshold of water content in the qualified oil condition range. The component outlet pressure in the re-inspection lockout judgment data is compared with the preset minimum start-up outlet pressure threshold. When the oil particle concentration is lower than the upper limit threshold for particle concentration, the oil moisture content is lower than the upper limit threshold for moisture content, and the component outlet pressure is higher than the minimum start-up outlet pressure threshold, the result indicating that the oil condition has been restored during the re-inspection is output; otherwise, the periodic re-inspection timer is reset, and the system waits for the next re-inspection test.

Images