A kind of anti-blocking control method and intelligent anti-blocking system of sampler
By deploying heterogeneous sensor arrays and a hybrid intelligent diagnostic architecture in key parts of the sampler, the problems of the singleness and lag of the sampler's blockage identification and clearing strategies are solved, enabling the sampler to operate efficiently and stably under complex conditions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUANENG POWER INT INC YINGKOU POWER PLANT
- Filing Date
- 2026-04-03
- Publication Date
- 2026-07-14
Smart Images
Figure CN122386658A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of industrial automation control technology, specifically relating to a sampling machine anti-blocking control method and intelligent anti-blocking system. Background Technology
[0002] As a key piece of equipment in industrial production for the automated and representative sampling of solid bulk materials (such as coal, ore, and grain), the continuity and reliability of the sampling machine's operation directly affect the fairness of subsequent quality analysis, process control, and trade settlement. During long-term operation, the internal material conveying path of the sampling machine (including the sampling head, chute, crusher, and divider) is highly susceptible to blockage due to factors such as material characteristics (e.g., high humidity, high viscosity, uneven particle size), environmental conditions (e.g., low temperature, high dust), and equipment structure. Blockage not only leads to sampling interruptions and reduced equipment utilization, but can also cause equipment overload damage, sample contamination, or loss, severely impacting production continuity and data accuracy.
[0003] In existing technologies, the main anti-blocking measures for sampling machines include the following:
[0004] Mechanical unblocking versus manual inspection: This method relies on regular manual inspections or the installation of mechanical unblocking devices (such as baffles, chains, etc.) at pre-set locations to intervene when blockages are detected or anticipated. This approach has a delayed response, cannot prevent blockages from occurring, and manual intervention poses safety risks and may compromise the representativeness of the sample.
[0005] Single signal threshold alarm: This method involves installing level switches, pressure switches, or monitoring motor current at key points. When the signal exceeds a preset threshold, an alarm or simple action (such as starting a rapper) is triggered. This method has a single monitoring dimension, cannot accurately distinguish the type and degree of blockage, has a high false alarm and false negative rate, and the blockage clearing method is crude, which may cause unnecessary impact on equipment or sampling processes.
[0006] Timed or sequential control for unblocking: This method uses timers or is interlocked with the equipment to periodically activate unblocking devices such as rapping and air blowing. However, this method lacks specificity, resulting in energy waste and equipment wear when there is no blockage, and may be ineffective in clearing blockages due to improper intensity or timing when there is severe blockage.
[0007] Control based on simple logic: This method combines a small number of sensor signals and triggers unblocking through simple logical judgments (such as "AND" and "OR"). This type of method has poor adaptability to complex and ever-changing blockage conditions, cannot handle contradictory and ambiguous information in multi-source signals, and has a low level of intelligence.
[0008] In summary, existing technologies generally suffer from problems such as limited monitoring methods, isolated information, crude diagnosis, delayed response, and a lack of adaptability and precision in clearing blockage strategies. This leads to the sampler's anti-blockage control often being in a passive state of "post-event processing" or "blind prevention," making it difficult to achieve the transformation from "fault handling" to "condition prevention," and failing to guarantee the efficient, stable, and reliable operation of the sampler under complex working conditions.
[0009] Therefore, there is an urgent need for an intelligent anti-blocking method and system that can perceive the operating status in real time and comprehensively, intelligently identify the risk and type of blockage, and adaptively execute precise and collaborative blockage clearing strategies. Summary of the Invention
[0010] This application provides a sampling machine anti-blocking control method and an intelligent anti-blocking system, aiming to solve the problems of existing technologies such as single monitoring methods, isolated information, crude diagnosis, delayed response, and lack of adaptability and accuracy in clearing strategies.
[0011] Firstly, a method for preventing blockage in a sampling machine, the method comprising:
[0012] S1: Real-time monitoring and data acquisition. A heterogeneous multi-sensor array is deployed in the key parts of the sampler that are prone to clogging to collect multi-source operating status data, including material flow rate, equipment vibration, motor current, internal pressure and images, and transmit the collected data to the central processing unit.
[0013] S2: Intelligent analysis and risk assessment of blockage characteristics. It receives multi-source operating status data collected by S1, analyzes it through a pre-trained multimodal data fusion diagnostic model, and outputs the blockage confidence, risk level and blockage type judgment for each monitoring point.
[0014] S3: Adaptive congestion clearing strategy decision and execution. Based on the congestion confidence and risk level output by S2, it calls the preset strategy decision tree to generate and execute the corresponding congestion clearing control strategy.
[0015] S4: Effectiveness evaluation and strategy optimization. After the congestion clearing action is executed, key parameters are collected and compared with the state before the congestion clearing, the congestion clearing efficiency index is calculated, and the data of this congestion event is stored in the historical database for the purpose of optimizing the diagnostic model and congestion clearing strategy.
[0016] Optionally, in S2, the multimodal data fusion diagnostic model adopts a hybrid architecture that combines a fuzzy logic preprocessing unit with a deep neural network classification unit.
[0017] The fuzzy logic preprocessing unit performs fuzzification and rule reasoning on the feature quantities from S1 based on a preset expert rule base, and outputs a fuzzy feature vector representing the tendency of the blocking type.
[0018] The deep neural network classification unit receives and fuses the fuzzy feature vector and the preprocessed original feature vector, performs analysis and processing, and outputs the final blockage confidence, risk level, and probability distribution of blockage type.
[0019] Optionally, step S3 includes: when the congestion confidence level is lower than the first threshold, the system operates in prevention mode, performing only data monitoring and trend analysis;
[0020] When the congestion confidence level reaches or exceeds the first threshold but is below the second threshold, the system enters the early warning intervention mode and performs non-destructive intervention actions.
[0021] When the congestion confidence level reaches or exceeds the second threshold, the system enters the active congestion clearing mode and performs the corresponding physical congestion clearing action according to the congestion type determined by S2.
[0022] Optionally, in the active unblocking mode, at least one of the following actions is performed according to the blockage type:
[0023] For adhesive blockages, activate a high-frequency micro-amplitude pneumatic oscillator;
[0024] For bridging blockages, activate the directional low-frequency high-energy pneumatic impactor;
[0025] For blockages caused by debris, control the micro-mechanical probe mechanism to probe or move it.
[0026] Optionally, in S4, the system runs a closed-loop learning and optimization engine, periodically using event records in the historical database to incrementally learn or retrain the diagnostic model in S2, and performs statistical analysis and optimization adjustment on the congestion clearing strategy parameters in S3 to improve congestion clearing efficiency.
[0027] Secondly, an intelligent anti-blocking system for a sampling machine includes: a distributed monitoring module deployed at the key blockage-prone parts of the sampling machine to collect multi-source operating status data;
[0028] The intelligent analysis and control module is connected to the distributed monitoring module, receives and analyzes the operating status data, outputs the blockage confidence level, risk level and blockage type judgment, and generates blockage clearing strategy instructions;
[0029] The multimodal execution unblocking module is connected to the intelligent analysis and control module, receives and executes the unblocking strategy instructions, and performs corresponding physical unblocking actions.
[0030] The intelligent analysis and control module also includes a data storage and management unit, which stores real-time operating data, historical event records and model parameters, and supports a closed-loop learning and optimization engine to optimize the diagnostic model and congestion clearing strategy.
[0031] Optionally, the distributed monitoring module includes: a material flow rate monitoring unit, a mechanical load and status monitoring unit, an internal pressure monitoring unit, a visual status monitoring unit, and an environmental condition monitoring unit; each unit is connected to the intelligent analysis and control module via a real-time industrial Ethernet fieldbus.
[0032] Optionally, the intelligent analysis and control module is deployed in an industrial-grade edge computing controller, and its software architecture includes: a data access and preprocessing unit, which receives and preprocesses data from the distributed monitoring module;
[0033] The multimodal congestion diagnosis algorithm unit uses a fuzzy neural network hybrid model for real-time diagnosis;
[0034] The adaptive control strategy generation unit calls the strategy library and generates clearing control commands based on the diagnostic results.
[0035] Optionally, the multimodal unblocking module includes at least one of the following: a micro-pulse airflow generator array, a high-frequency micro-amplitude aerodynamic oscillator array, a directional low-frequency high-energy aerodynamic impactor, and a micro-mechanical probe mechanism.
[0036] Optionally, it also includes: a sampling process coordinator, which communicates and interlocks with the main control system of the sampling machine during the execution of the unblocking action to coordinate the unblocking action and the sampling process;
[0037] The human-computer interaction and remote operation and maintenance module provides local status display, alarm, and manual operation interface, and supports remote data transmission, configuration and diagnostic functions.
[0038] Compared with the prior art, this application has at least the following beneficial effects:
[0039] This application constructs a sensing network covering the entire material flow path by deploying heterogeneous multi-sensor arrays (including multi-dimensional sensors such as flow, vibration, current, pressure, image, and environment) at multiple key locations prone to blockage. This network can capture early and weak characteristic signals of blockage from multiple physical levels. The high-frequency and synchronized data acquisition provides a high-quality data foundation for subsequent in-depth analysis, overcoming the one-sidedness and lag of single signal monitoring.
[0040] This application employs a hybrid intelligent diagnostic architecture that combines a fuzzy logic preprocessing unit and a deep neural network classification unit. The fuzzy logic preprocessing unit effectively utilizes domain expert knowledge to transform multi-source continuous features into semantically rich fuzzy feature vectors, providing interpretable prior guidance for diagnosis. The deep neural network classification unit, through deep fusion analysis of the original features and fuzzy feature vectors, learns the deep nonlinear relationship between multi-source information and blockage status under complex working conditions. This synergy at the feature level achieves a complementary advantage between symbolic knowledge and connectionist learning capabilities, enabling the system to not only output high-precision blockage confidence and risk levels but also effectively distinguish blockage types, providing crucial and reliable decision-making basis for subsequent precise blockage clearing.
[0041] Based on real-time diagnostic results, this application employs a tiered control strategy (prevention, early warning, and proactive) triggered by thresholds. In early warning mode, flexible interventions (such as parameter fine-tuning and minor airflow disturbances) aim to prevent material blockage from developing, thus preventing problems before they occur. Proactive mode, on the other hand, invokes targeted physical unblocking mechanisms (such as high-frequency oscillators, directional impactors, and mechanical probes) based on the type of blockage, achieving a "targeted approach" and avoiding the blind and destructive nature of unblocking actions. This efficiently unblocks while minimizing interference with the sampling process and sample quality. Attached Figure Description
[0042] Figure 1 A flowchart illustrating an anti-blocking control method for a sampling machine, provided as an embodiment of this application;
[0043] Figure 2 This is a schematic diagram of the module connection of an intelligent anti-blocking system for a sampling machine provided in one embodiment of this application. Detailed Implementation
[0044] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figure 1-2 The present application will be further described in detail with reference to the embodiments.
[0045] This application provides a method for preventing blockage in a sampling machine, comprising the following steps:
[0046] S1: Real-time monitoring and data acquisition. Based on modularity and function-oriented principles, a heterogeneous multi-sensor array is deployed at the key, easily clogged parts of the sampler to construct a sensing network covering the entire material flow path. The key parts, defined according to the material flow and clogging mechanism, include at least: the interior of the sampling head drilling cylinder and the material inlet contour area; the primary inclined chute connecting the sampling head and the crushing unit; the crusher inlet and key cross-sections of its internal cavity; the rotator or cutter cavity for sample reduction; and the secondary chute that transports the final sample to the collection container.
[0047] Specifically, the monitoring implementation method is as follows: Multiple capacitive or radio frequency admittance level switches are arranged in a ring around the sampling head inlet profile to detect whether material overflow or top bulging occurs inside the drilling barrel; a miniaturized piezoresistive pressure sensor is installed near the drill bit inside the drilling barrel to monitor the real-time resistance of the material to the drill bit during drilling. Non-contact microwave Doppler flow meters are installed in the initial, middle, and contraction sections of the primary chute, respectively, to continuously measure the velocity profile of the surface material; simultaneously, triaxial piezoelectric accelerometers are installed at corresponding positions on the outer side of the chute wall, with rigidly reinforced mounting bases to accurately transmit inner wall vibration signals. A Hall effect current sensor is coupled and installed in the main circuit of the crusher's drive motor to acquire the three-phase current waveform of the motor in real time; industrial-grade vibration accelerometers are installed at the crusher housing bearing seats and the feed inlet flange to acquire wideband (e.g., 0-10kHz) vibration time-domain signals. Inside the divider chamber, dynamic pressure diaphragm sensors are installed at the fulcrums of the rotating grid or oscillating chute and at the material impact points to capture transient pressure pulsations as the material passes through. At the outlet of the secondary chute, a dustproof industrial endoscope is installed, its lens equipped with an air curtain protection device, to periodically or trigger-based acquire visible light and near-infrared images of the chute's interior. Furthermore, digital temperature and humidity sensors are installed inside the sampling head's protective cover and at the crusher's ventilation openings to monitor ambient temperature and humidity parameters.
[0048] All sensor output signals, including analog current / voltage signals, digital pulse signals, and image data streams, are transmitted to the central processing unit via a fieldbus network (such as PROFIBUS-DP or EtherCAT). The sampling frequency for key dynamic signals such as vibration and current is no less than 10kHz to meet the frequency resolution requirements of subsequent spectrum analysis. The sampling period for process parameters such as flow rate and pressure is configurable, but should not exceed 100ms during normal operation. To cope with harsh field environments (dust, vibration, electromagnetic interference), all sensor selections must meet the corresponding protection level (no less than IP65) and vibration resistance. Signal transmission uses shielded twisted-pair cables or optical fibers, and signal isolation and conditioning modules are configured at the access end to ensure the reliability and accuracy of data acquisition. The data acquisition system has a self-diagnostic function, capable of reporting sensor disconnection, signal over-limit, and other fault states in real time. The collected raw data, synchronized with timestamps and location tags, forms standardized real-time operating status data blocks for subsequent analysis modules to access.
[0049] S2: Intelligent Analysis and Risk Assessment of Congestion Features. The multi-source heterogeneous real-time operating status data collected by S1 is synchronously input into an embedded intelligent analysis module. This module is deployed within an industrial-grade edge computing gateway, and its core is a pre-trained multimodal data fusion diagnostic model. This model adopts a hybrid architecture that combines a fuzzy logic preprocessing unit and a deep neural network classification unit to achieve real-time analysis and risk assessment of congestion features.
[0050] Specifically, the intelligent analysis module operates according to the following hierarchical process:
[0051] Data preprocessing layer: First, the raw data of each channel is standardized, filtered for noise reduction, and time-aligned. For vibration signals, a combination of Fast Fourier Transform and wavelet packet decomposition is used to extract features such as energy and spectral kurtosis within a specific frequency band; for motor current signals, Park vector transform is performed to obtain amplitude, phase, and harmonic features; for pressure and flow signals, their rate of change, statistical characteristics, and approximate entropy are calculated; for image data, texture features such as contrast, correlation, and entropy are extracted. After preprocessing, a set of multi-dimensional raw feature vectors is obtained.
[0052] Fuzzy Logic Preprocessing Unit: This unit acts as a feature enhancer based on expert knowledge. It fuzzifies the key continuous features (such as "flow rate decrease" and "current harmonic distortion rate") in the original feature vector using predefined membership functions (such as triangular or Gaussian functions), transforming them into semantic variables such as "low," "medium," and "high." Subsequently, parallel reasoning is performed based on a pre-defined IF-THEN expert rule base (e.g., "IF Flow rate decrease IS 'high' AND Pressure fluctuation entropy IS 'low' THEN Bridging tendency IS 'strong'"). The output of this unit is not a simple decision label, but a "fuzzy feature vector." This vector integrates information on the strength of tendency towards different blockage types (such as adhesion, bridging, and jamming) derived from multiple rules, and is a structured intermediate feature representation rich in domain prior knowledge.
[0053] Deep Neural Network Classification Unit: This unit is the core of the final intelligent classification decision. Its input layer simultaneously receives fuzzy feature vectors from the previous stage and filtered, normalized original feature vectors. In the hidden layers of the network, these two feature representations from different sources and with complementary information undergo deep fusion and interactive computation. Through its nonlinear transformation capabilities, the network learns the complex mapping relationship from the fused information of "raw data features" and "expert knowledge features" to the final congestion state;
[0054] Comprehensive evaluation output: After processing by the deep neural network classification unit, the final comprehensive decision result is generated at the output layer: a) Blockage confidence: a scalar between 0 and 1, representing the overall probability of effective blockage; b) Risk level: such as three-level classification of "normal", "warning", and "alarm"; c) Blockage type probability distribution: a vector, representing the probability of belonging to each preset blockage type (e.g., [adhesion: 0.75, bridging: 0.20, jamming: 0.05]).
[0055] The embedded intelligent analysis module supports online fine-tuning and incremental learning of model parameters, and can continuously optimize diagnostic rules and network weights based on actual on-site operation feedback data.
[0056] S3: Adaptive congestion clearing strategy decision and execution. Based on the real-time output of S2 of the congestion confidence level (C, range 0-1) and risk level (R, e.g., N1-normal, N2-warning, N3-alarm) of each monitoring point, the central control unit calls the preset strategy decision tree to adaptively trigger and execute a hierarchical and progressive congestion clearing control strategy. The first threshold ( ) and the second threshold ( The threshold value is a dynamic threshold set based on historical data and experimental calibration. Preferably, The value range is 0.4-0.6. The value range is 0.7-0.8.
[0057] When the system determines that C at all monitoring points is less than When R is N1, the system operates in prevention mode. In this mode, no physical clearing actions are initiated; the system focuses on high-frequency data acquisition, storage, and trend analysis. The control unit continuously calculates the moving average and standard deviation of key parameters (such as average flow velocity and average vibration energy) and performs short-term predictions based on time series models (such as autoregressive models), providing early insights into potential risk evolution. All data is recorded in a timestamped circular database for model optimization and incident tracing.
[0058] When C at any monitoring point reaches or exceeds But smaller than When the R value jumps to N2, the system immediately triggers an early warning intervention mode. This mode aims to change the material flow pattern through non-destructive, flexible intervention to prevent further blockage. Specific execution strategies include: 1) Fine-tuning of process parameters: By adjusting the preset values in the sampler PLC, the cutting angle of the sampling head is dynamically increased by 1° to 3° from the original setting, or the horizontal running speed of the sampling arm is increased by 5% to 15% from the rated value, thereby changing the initial kinetic energy and trajectory of the material entering the subsequent process. 2) Active airflow disturbance: The micro-pulse airflow generator located upstream of the target monitoring point is activated. The generator is controlled by a high-speed solenoid valve, and its operating parameters are optimized as follows: pulse pressure 0.05-0.15MPa, pulse duration 50-200 milliseconds, and pulse frequency 0.5-2Hz. This low-energy, intermittent airflow aims to "blow" and disturb the initial material layer that may form without significantly changing the overall material conveying volume, thus disrupting the initial adhesion tendency between particles.
[0059] When C at any monitoring point reaches or exceeds When R is N3, the system unconditionally switches to active blockage clearing mode. Before this mode is activated, the system first integrates the fuzzy logic and neural network's judgment suggestions on the blockage type (adhesive type, bridging type, or debris blockage type) in step S2, and combines the specific location information of the blockage point to call up the preset, targeted blockage clearing combination strategy;
[0060] For clogging diagnosed as adhesive blockage (characterized by a slow decrease in flow rate, increased high-frequency vibration components, and thickened material layer on the wall surface as shown in images), the system activates a high-frequency micro-amplitude pneumatic oscillator installed on the outer side of the wall panel corresponding to the blockage point. This oscillator employs a piston design, with its operating parameters set as follows: operating pressure 0.3-0.6 MPa, impact frequency 20-60 Hz, and duration of each action 3-10 seconds. The low-amplitude, high-frequency mechanical vibrations generated are directly transmitted through the wall panel to the adhesive layer on the inner wall, breaking down the bonding force between the material and the wall surface, as well as between material particles, through fatigue effects, causing it to peel off.
[0061] For bridging-type blockages (characterized by a sudden drop or interruption of flow velocity, a significant decrease in pressure fluctuation entropy, and abnormal vibration signals at specific locations), the system activates a directional low-frequency high-energy pneumatic impactor installed directly above or to the side of the material arch formation area. The impactor operates by rapidly releasing energy from the storage chamber, generating a powerful pneumatic hammer effect with single or low-frequency bursts (1-5 Hz). Its key parameters are: impact pressure 0.7-1.2 MPa, and single impact energy 50-200 joules. The generated directional stress wave directly acts on the critical stress points of the material arch structure, using instantaneous high-energy input to disrupt its mechanical equilibrium, causing the arch structure to collapse.
[0062] For the initial judgment of debris blockage type (the characteristics may be manifested as the motor current surging and then getting stuck at a specific point, or irregular contour foreign objects being identified by image recognition), the system first attempts to conduct a probe through short-term and intermittent reverse airflows or slight vibrations. If it is ineffective, it controls the action of the micro-mechanical probe mechanism deployed near the key passes (such as the inlet of the splitter grid). This mechanism includes a retractable probe driven by a servo motor with a blunt or hook-shaped front end. Under the control instruction, the probe extends into the suspected blockage area with a preset force and displacement mode (for example, first slowly extend 50 mm with a force of 5 - 10 N, and then reciprocate slightly 3 - 5 times at a frequency of 2 - 5 Hz) to conduct mechanical probing,拨动 or hooking, in order to loosen or remove the stuck debris;
[0063] During the execution of any physical blockage clearing action in the active blockage clearing mode, the system conducts real-time communication and interlock control with the main control system of the sampler through the sampling process coordinator to ensure that the core logic of the sampling operation is not damaged. The coordination strategies include: 1) For the blockage clearing actions occurring in the core processes such as sampling, crushing, and splitting, the system can request the main control system to briefly pause the feeding or conveying link of the current sampling cycle (the pause duration is usually no more than 30 seconds), and at the same time freeze the timer, and seamlessly resume after the blockage clearing action is completed, so as to ensure that the material integrity of this sampling is not contaminated by the unblocking airflows or vibrations that interrupt it; 2) If the equipment is designed with a standby bypass chute, during the early warning or blockage clearing period, part of the material flow can be briefly switched to the bypass to ensure the continuous and stable feeding of the main line sampling process. The execution process and results of all blockage clearing actions (such as the confirmation signal of the restored flow) are monitored in real time and input as feedback information into S4;
[0064] S4: Effect evaluation and strategy optimization: After each blockage clearing action is triggered and executed, the system immediately enters the effect evaluation stage. In this stage, taking the end moment of the blockage clearing instruction as the time origin (T0), an evaluation window with a preset duration (such as 60 seconds) is started. Within this window, the system continuously and highly frequently collects and analyzes the key parameters described in S1, especially the material flow rate, vibration spectrum, motor current, and pressure signals at the target blockage point, and compares them with the baseline state before the blockage clearing (before T0) and the expected values under the theoretically unblocked state;
[0065] The quantitative evaluation of the blockage clearing effect is achieved by calculating multiple indicators: 1) Response time constant : Through the exponential fitting method, calculate the time required for the key parameter (such as the flow rate) to recover from the abnormal value to 63.2% of the steady state value, so as to evaluate the response speed of the blockage clearing action. 2) Recovery degree : It is defined as the percentage of the actual value of the key parameter to the theoretically unblocked expected value at the end of the evaluation window. 3) Disturbance index Assessing the negative impact of the unblocking action itself (such as impact and vibration) on other parts of the sampling process (such as sampling accuracy and reduction ratio stability) can be done indirectly by simultaneously monitoring instantaneous fluctuations in sample collection volume or analyzing vibration changes in secondary chutes. The final unblocking efficiency... It is a comprehensive score, calculated using the formula. The calculation shows that k1, k2, and k3 are weighting coefficients. This is a scoring function based on the time constant;
[0066] Subsequently, the system constructs a complete event record for this congestion event and stores it in a dedicated historical database. Each record includes, but is not limited to, the following fields: unique event ID, timestamp, location identifier of the congestion, trigger mode (early warning / active), multi-dimensional feature data vector before clearing the congestion (i.e., a snapshot of the congestion features), the congestion type judgment and confidence level output in step S2, the specific set of clearing strategy instructions executed (including the type, location, and operating parameters of each actuator, such as pressure, frequency, and duration), the sensor data sequence during the clearing process, and the calculated clearing efficiency η and its sub-indicators. All records are indexed by time and can be retrieved using tags such as location, congestion type, and clearing strategy.
[0067] The system runs a low-priority, controlled, closed-loop learning and optimization engine in the background. The core design principles of this engine are "data-driven optimization, simulation-based security verification, and controlled incremental deployment" to ensure that all updates improve performance without introducing systemic risks or disrupting the stability of the sampling process.
[0068] Its safe and reliable workflow is as follows:
[0069] Update Triggering and Data Preparation: The engine does not run continuously, but starts when preset safety trigger conditions are met (e.g., every N accumulated valid congestion event records; or every T hours during a system idle maintenance period). After starting, the engine selectively extracts event records from the historical database that meet data quality standards within a certain period to form an offline training and validation dataset. Data selection excludes events under abnormal conditions (such as during sensor malfunctions) or events that have not been evaluated, ensuring a reliable learning foundation.
[0070] Offline analysis and candidate update generation: All core calculations and update package generation are performed in the system's offline sandbox environment or the non-real-time security partition of the edge controller, isolated from the online real-time control system;
[0071] For the diagnostic model (deep neural network part): The engine uses the new dataset to incrementally learn or retrain the existing model, generating one or more candidate model update packages. Each candidate package is accompanied by a performance evaluation report on the validation set (such as improvements in accuracy and recall).
[0072] For fuzzy logic rule bases: the engine analyzes the correlation between features and final blocking types in event logs, automatically calculates and proposes adjustments to membership function parameters or rule weights, but does not directly modify them. These suggestions are presented in the form of a configuration change suggestion list;
[0073] For the congestion clearing strategy library: For each type of congestion at a specific location, the engine uses statistical analysis (such as regression analysis) or optimization algorithms (such as Bayesian optimization) to find the relationship between strategy parameters (such as pressure P, time T) and congestion clearing efficiency. The correlation between them is determined, and optimized candidate values for strategy parameters are generated.
[0074] Security verification and confirmation: Any generated candidate updates (models, parameters, rules) must undergo rigorous multi-level verification before entering the deployment process.
[0075] Simulation verification: First, in the digital twin simulation system, historical data playback or simulated data are used to verify whether the behavior of candidate updates under various operating conditions meets expectations, ensuring that there are no logical conflicts or performance degradation.
[0076] Small-scale security testing: After authorization (e.g., confirmation by an engineer), candidate updates can be applied to a single device or an isolated subsystem and run briefly in monitoring mode. In this mode, the system runs the old and new logics in parallel and compares the output results, but does not actually execute the control commands generated by the new logic, thereby evaluating the rationality of its decision-making in a real-world environment;
[0077] Engineer Review and Confirmation: All adjustment suggestions, performance reports, and security test results are provided to maintenance engineers through the human-computer interaction module. Critical changes to logic rules, model replacements, and significant adjustments to core strategy parameters must be reviewed and manually authorized by the engineer before proceeding to the next step.
[0078] Controlled deployment and effect monitoring: Updates are verified and confirmed, and deployed in the form of versioned update packages within the maintenance window set by the system.
[0079] The deployment process is reversible, and the system retains the function of rolling back to the previous stable version with one click;
[0080] After the update, the system enters a pre-defined enhanced monitoring period. During this period, the performance metrics of the updated components are recorded and their effects are tracked more intensively, and compared with the baseline data before the update to confirm the stability of the optimization effect.
[0081] If any unexpected negative effects occur during the monitoring period (such as an increase in false alarm rate or abnormal clearing actions), the system will automatically alarm and trigger a preset rollback procedure to restore to the previous stable state.
[0082] Through the above process, the system forms a complete intelligent closed loop of "assessment-learning-verification-confirmation-deployment-monitoring". While fully utilizing the value of data for self-optimization, this closed loop ensures the extremely high reliability and operational safety required in industrial settings through multiple security boundaries and human supervision, enabling the system to achieve continuous performance improvement safely and robustly.
[0083] In one embodiment, an intelligent anti-blocking system for a sampling machine is also provided, comprising:
[0084] The distributed monitoring module collects operational status data at key, easily clogged parts of the sampler;
[0085] The distributed monitoring module, serving as the system's sensing layer, employs redundant configuration and functional partitioning principles, deploying heterogeneous sensor arrays at various critical, easily clogged locations along the material path of the sampling machine. The physical structure and specific implementation of this module are as follows:
[0086] Material flow rate monitoring unit: Non-contact microwave Doppler flow meters are embedded in the initial, middle, and contraction sections of the primary chute, and at the inlet and outlet of the secondary chute. These flow meters preferably operate in the 24GHz or 80GHz frequency band, and their antenna emission angle is precisely designed to form a fan-shaped beam that matches the chute's cross-sectional width, measuring the average flow velocity and velocity profile of the surface material. During installation, the flow meter probe passes through a custom flange on the chute wall, with its front-end emission surface flush with or slightly recessed (no more than 2mm) to avoid forming material accumulation protrusions. To compensate for the influence of changes in the material's dielectric constant, the unit incorporates a self-calibration algorithm that periodically corrects the measurement reference based on reference operating conditions.
[0087] Mechanical load and condition monitoring unit: This unit consists of a vibration monitoring subunit and a current monitoring subunit;
[0088] Vibration monitoring subunit: Triaxial piezoelectric accelerometers are rigidly mounted on the slewing bearing housing of the sampling head, the bearing housings at the drive and non-drive ends of the crusher, and at the reinforcing ribs or flange connections of each key chute. The sensor mounting base is made of stainless steel and is rigidly connected to the housing using epoxy resin adhesive or double-ended bolts. Its resonant frequency is much higher than the analysis frequency band of interest (typically >10kHz). The sensor measurement range must cover ±50g or more, with a frequency response of not less than 5kHz, for acquiring broadband vibration acceleration signals.
[0089] Current monitoring subunit: Closed-loop Hall effect current sensors are coupled and installed on the main circuits of each phase within the power cabinets of the sampling head lifting motor, crusher drive motor, and reducer drive motor. The sensor output signal is an analog voltage signal proportional to the primary current (e.g., ±10V corresponds to ±100A), and its bandwidth (typically >5kHz) must be sufficient to capture transient current distortion and harmonic components caused by stall or jamming.
[0090] Internal pressure monitoring unit: A miniature diffused silicon piezoresistive pressure transmitter is installed in locations prone to forming enclosed material sections or pressure accumulation, such as inside the sampling head drill barrel near the drill bit, at the top and sides of the crushing chamber, and at the end of a long horizontal chute. This transmitter uses a fully welded stainless steel diaphragm structure, with a process connection of G1 / 4 or M20×1.5 thread. Its measurement range is selected according to the location, for example, 0-1MPa inside the drill barrel and 0-50kPa inside the chute. The signal output is a 4-20mA analog signal, with built-in temperature compensation to improve long-term stability.
[0091] Visual Status Monitoring Unit: Dustproof and waterproof industrial endoscopes are installed in easily observable locations with controllable lighting conditions, such as the divider cavity and the transition point of the secondary chute. This endoscope integrates a camera, LED lighting module, air curtain protection interface, and optional mechanical swing mechanism within a cylindrical protective housing. The protective housing has a rating of at least IP67, and the front end is constantly vented with clean, dry, low-pressure air (approximately 0.1-0.3 MPa) to form an air curtain, preventing dust from adhering to the lens. The camera resolution is at least 2 megapixels and can output JPEG images or low frame rate video streams.
[0092] Environmental Condition Monitoring Unit: Digital temperature and humidity sensors (such as those based on capacitive polymer humidity sensors) are distributed throughout the sampling head housing, near the crusher air inlet, and within the electrical control cabinet. These sensors directly output calibrated temperature and relative humidity values via digital interfaces such as I2C or Modbus RTU, with a typical measurement accuracy of ±0.5°C for temperature and ±3%RH for humidity.
[0093] All the aforementioned sensors are connected to nearby distributed field data acquisition stations via shielded twisted-pair cables or industrial Ethernet cables. Each acquisition station is equipped with multi-channel analog input modules, digital input modules, and a dedicated image acquisition card, responsible for analog-to-digital conversion, preliminary filtering, and format standardization of the raw signals. The acquisition stations are interconnected with each other and with the central controller via a real-time industrial Ethernet fieldbus (preferably EtherCAT or PROFINETIRT), forming a deterministic distributed data acquisition network. The network communication cycle can be configured as needed, typically 1-10 milliseconds, ensuring synchronization and low latency of status data. Each acquisition station and sensor has independent power and communication status diagnostic functions, enabling module-level fault alarms and location.
[0094] The intelligent analysis and control module is connected to the distributed monitoring module, receives and analyzes data, outputs the congestion confidence level, risk level and type judgment, and generates congestion clearing strategy instructions;
[0095] The intelligent analysis and control module employs a highly reliable embedded hardware platform and a layered software architecture to achieve fusion analysis of multi-source monitoring data, intelligent diagnosis of congestion status, and real-time generation of congestion clearing strategies. The intelligent analysis and control module includes:
[0096] The hardware platform for this module is an industrial-grade edge computing controller, whose core is a high-performance multi-core embedded processor (such as a heterogeneous multi-core SoC based on the ARM Cortex-A and Cortex-M series). This processor primarily handles algorithm computation and logic control tasks. The system software runs on a real-time operating system (RTOS) to ensure deterministic task scheduling and low latency. The software functionality is divided into the following core units:
[0097] Data Access and Preprocessing Unit: This unit receives raw data packets from each acquisition station of the distributed monitoring module in real time via a fieldbus interface (such as an EtherCAT slave controller). First, it performs data unpacking, timestamp alignment, and channel matching. Then, it performs targeted preprocessing for different types of signals: vibration and current signals are resampled to a unified analysis frequency and anti-aliasing filtering and detrending processing are applied; pressure and flow signals are subjected to moving average filtering to suppress impulse interference; and image data is decoded and format standardized. The preprocessed data is placed in a shared real-time data buffer.
[0098] Multimodal congestion diagnosis algorithm unit: Employs a cascaded fuzzy neural network hybrid model. This model runs on the processor's application processing core and is triggered to execute once per control cycle (e.g., 100ms).
[0099] Front-end fuzzy feature extractor: First, key features are extracted from the buffer, such as "flow rate decrease rate," "current harmonic distortion rate," "specific frequency band vibration energy ratio," and "pressure approximate entropy." Each feature is fuzzified using a set of predefined, configurable membership functions (e.g., π-type or sigmoid functions), transforming it into semantic variables such as "very low," "low," "medium," "high," and "very high." A fuzzy inference engine based on an expert knowledge base performs parallel inference according to preset rules (e.g., "IF flow rate decrease rate IS 'high' AND pressure approximate entropy IS 'low' THEN bridging probability IS 'high'"), outputting a preliminary probability distribution (a fuzzy set) for several basic blockage types (adhesion, bridging, jamming).
[0100] The backend deep neural network classifier concatenates the fuzzified feature vector (including the original feature values and their fuzzified semantic variable encodings) with the probability distribution vector output by the frontend fuzzy inference, feeding both into a deep neural network as input. The preferred network structure is a multilayer perceptron with several fully connected hidden layers, or a combination of one-dimensional convolutional layers and fully connected layers suitable for time series analysis. The network output layer uses the Softmax function, ultimately outputting a multi-dimensional vector containing: a) Blockage confidence: a scalar between 0 and 1, representing the overall probability of effective blockage; b) Risk level: a classification output (e.g., 0-normal, 1-warning, 2-alarm); c) Blockage type probability distribution: a vector representing the probability of belonging to each preset blockage type. This neural network model requires sufficient offline and online training on laboratory and field historical data, with the weight file stored in non-volatile memory.
[0101] Adaptive Control Strategy Generation Unit: This unit runs on the processor's real-time control core and receives the output results from the diagnostic algorithm unit each cycle. It embeds a strategy decision state machine and a parameterized strategy library. The decision state machine determines the appropriate control mode (prevention, warning, or active) based on the current confidence level, risk level, and primary suspected blockage type, combined with the equipment's current operating mode (e.g., sampling in progress, intermittent operation). Once the mode and strategy type are determined, the unit calls the corresponding strategy template from the strategy library. The strategy template is an abstract description of executable instructions, containing actuator identifiers, action types, and key parameters (e.g., pressure setpoint P, frequency F, duration T). Based on the specific location and severity (confidence level) of the blockage, the unit uses built-in interpolation algorithms or empirical formulas to calculate the specific values of these parameter variables in real time, thereby generating a specific, deployable blockage-clearing control command sequence. This unit is also responsible for safety interlock communication with the sampling machine's main control PLC to coordinate blockage-clearing actions with the production process.
[0102] Data storage and management unit: This unit manages local non-volatile storage media (such as eMMC or solid-state drives). The storage content is divided into three main parts:
[0103] Real-time running database: Stores raw and preprocessed data from all sensors within a recent period (e.g., 24 hours) in a circular buffer for real-time trend display and short-term backtracking analysis;
[0104] Historical event database: Stores each complete blockage event record in a structured table format, including event feature snapshots, diagnostic results, execution strategies, process data, effect evaluation results, etc., and supports querying by time, location, type and other conditions;
[0105] Model and Configuration Library: Stores neural network model weight files, membership function parameters and rule bases for fuzzy logic, all templates and parameter tables for the policy library, and system configuration parameters. This unit provides a controlled access interface for the closed-loop learning optimization engine described in S4 to read model and event data and write updated model and policy parameters.
[0106] The modules exchange data between their internal units via an internal message bus or shared memory, and are equipped with a hardware watchdog and software heartbeat monitoring to ensure long-term stable operation in harsh industrial environments. Externally, the modules provide standard industrial Ethernet interfaces and configuration / debugging interfaces, supporting remote status monitoring, parameter configuration, and firmware upgrades.
[0107] The multimodal execution unblocking module is connected to the intelligent analysis and control module, receives and executes unblocking strategy instructions, and performs physical unblocking actions.
[0108] The multimodal blockage clearing module, serving as the system's physical execution layer, consists of a series of dedicated pneumatic and mechanical mechanisms. Controlled by a sequence of commands issued by the intelligent analysis and control module, it enables precise intervention for blockages of different types and locations. Specifically, this module includes:
[0109] Micro-pulse airflow generator array: This array consists of multiple independently controlled high-speed normally closed solenoid valves and their matching nozzles. The air inlet of each solenoid valve is connected to the system air source (pressure typically 0.4-0.8 MPa) via a centrally located clean air supply pipeline, and its outlet is connected to a specially designed stainless steel nozzle. The nozzle is embedded at a specific angle (typically 15° to 45° downwards) into the wall of the chute or pipe, with its outlet flush with or slightly recessed (≤1mm) to avoid obstructing material flow. The nozzle orifice diameter is optimized, for example, 1-2mm. Upon receiving a "warning intervention" command, the control module precisely controls the opening time (pulse width 50-200ms) and opening frequency (0.5-2Hz) of the corresponding solenoid valve, thereby generating a series of low-frequency, low-energy air pulses (small instantaneous flow rate, weak impact force) upstream of the target location. This is primarily used to disturb the surface of materials about to stagnate, preventing the formation of initial material accumulation, rather than for forceful purging.
[0110] High-frequency micro-amplitude pneumatic oscillator array: For wall areas prone to material adhesion (such as the bottom of chutes and contraction sections), a piston-type high-frequency pneumatic oscillator is rigidly mounted on the external back plate using high-strength bolts. The oscillator contains a piston driven by controlled airflow, which performs high-frequency reciprocating motion within a precision cylinder, transmitting the vibration to the mounting base via an impact anvil. Its key characteristics are: high impact frequency (20-60Hz), but low single impact energy (typically less than 5 joules) and small amplitude (typically less than 1mm). When adhesive blockage is diagnosed, the control module activates the air path of the oscillator at that location, driving it to operate for several seconds (e.g., 3-10 seconds). The resulting continuous high-frequency micro-amplitude vibration is transmitted through the wall plate, effectively breaking down the adhesion forces between material particles and the metal wall surface, as well as the cohesive forces between particles, causing fatigue and peeling of the adhesive layer, while minimizing impact on the overall structure.
[0111] Directional low-frequency high-energy pneumatic impactor: This device is specifically designed to break up dense material arches or compacted blockages. It is typically installed above abrupt changes in the cross-section of a chute, at the waist of a conical hopper, or other critical locations prone to forming mechanical arches. Essentially, it is a single-acting pneumatic hammer, comprising an inflatable energy storage chamber, a rapid-release valve (such as a rupture diaphragm or high-speed solenoid valve), and a relatively large impact piston. During operation, the energy storage chamber is first inflated to a predetermined high pressure (e.g., 0.7-1.2 MPa). Upon receiving an "active unblocking" command, the control module triggers the rapid-release valve, and the high-pressure gas instantly accelerates the piston, causing it to strike a force transmission rod rigidly connected to the equipment wall at a very high final velocity. This generates a strong, rapidly decaying directional stress wave within the wall. Its characteristics include high single-impact energy (50-200 Joules) but a low operating frequency (can be triggered in a single strike or in low-frequency continuous strikes of 1-5 Hz). The stress wave propagates through the material, effectively disrupting the force chain balance of the arched structure.
[0112] Miniature Mechanical Probe Mechanism: This mechanism is used to handle blockages caused by flexible debris or large foreign objects, and is typically integrated into the throat area such as the material inlet of the sampling head or the feed inlet of a crusher. It consists of a waterproof and dustproof servo electric cylinder, a planetary gear reducer, and a high-strength alloy probe with a blunt tip or barbs. The entire mechanism is encapsulated in a robust housing, with the probe initially retracted into the housing. When a suspected blockage is detected, the control module commands the servo motor to drive the probe, slowly extending it according to a preset program (adjustable speed, e.g., 5-20 mm / s), and performing short-stroke reciprocating insertion / removal or rotational agitation at the suspected blockage point (e.g., reciprocating stroke 10-30 mm, frequency 2-5 Hz). Its action is gentle yet possesses mechanical penetrating force, designed to pry open or hook out the stuck object, avoiding secondary damage to the equipment.
[0113] The sampling process coordinator is a hardware and software integrated logic control unit. On the hardware side, it is a dedicated communication interface within the intelligent analysis and control module (typically a network port or fieldbus interface card supporting mainstream industrial protocols such as PROFIBUS and ModbusTCP). On the software side, it runs a safety interlock logic program. This program incorporates a timing logic model for the normal operation of the sampler. When a clearing action is required, the coordinator first assesses the potential impact of this action on each stage of the current sampling cycle (such as sampling arm extension, drilling, lifting, discarding, reducing, and collecting samples). Then, it sends requests or direct commands to the sampler's main control PLC via the communication interface. For example, it requests the insertion of a brief waiting window in non-critical stages (such as the discarding or return process); or commands the temporary closure of the upstream feed valve during clearing; or switches the material flow when the equipment has redundant paths. Its core objective is to freeze or adjust non-core timing while allowing the physical clearing action to occur, ensuring that the representativeness and continuity of the material in each sample are not disrupted. All coordination commands and the response status of the main control system are recorded.
[0114] All actuators have integrated wiring for air supply, power supply and control signal lines, and are equipped with necessary safety valves, filter pressure reducing valves and status feedback sensors (such as those used to detect whether the impactor has been fired or whether the probe is in place).
[0115] The human-computer interaction and remote operation and maintenance module is connected to the intelligent analysis and control module, providing local status display, alarm and manual operation interface, and supporting remote data transmission, configuration and diagnosis;
[0116] The human-computer interaction and remote operation and maintenance module consists of two parts: a local interactive terminal and a remote communication gateway, as detailed below:
[0117] Local Human-Machine Interface Terminal: The main body of this terminal is an industrial-grade embedded touchscreen computer, installed on the door panel of the sampling machine's electrical control cabinet or on-site operation box. The screen size is typically no less than 10 inches, with a protection rating of no less than IP65 to withstand dusty and humid environments. It runs graphical monitoring software (HMI) specifically developed for the system. The software interface is based on a functional partition design, mainly including the following display and operation areas:
[0118] Comprehensive Status Overview Area: Dynamically displays the material path of the sampler in the form of a process flow diagram. Each monitoring point is color-coded (green - normal, yellow - warning, red - alarm) to reflect its blockage confidence and risk level in real time. Key parameters (such as main chute flow rate and crusher current) are displayed side-by-side in the form of digital instruments or trend curves.
[0119] Real-time Data and Alarm List Area: Displays real-time readings of all sensors in a scrolling table format. When the system triggers a warning or alarm, the alarm list automatically pops up, recording the alarm time, location, type (e.g., "Level 1 Chute Mid-section - Bridge Warning"), confidence level, and recommended measures. Alarms are sorted by priority and accompanied by audible and visual alerts.
[0120] Historical Data Tracking Area: Provides historical trend query function, allowing users to customize time periods and select parameters (up to 8 parameters can be displayed simultaneously) to view historical curves. Additionally, users can query the historical event database, viewing complete records of each congestion event by filtering by time, location, and type, including data changes before and after the event, the implemented congestion clearing strategies, and effectiveness evaluation reports.
[0121] Manual Operation and Mode Selection Area: In authorized mode, a manual intervention interface is provided. Operators can use this interface to: a) manually force start or stop any unblocking actuator (e.g., jog a pneumatic impactor) for testing or emergency handling; b) switch the system's automatic control mode (e.g., switch from "fully automatic" to "monitor only" or "maintenance mode"); c) temporarily modify key threshold parameters (e.g., warning thresholds). Modifications require password confirmation and are logged.
[0122] System Diagnosis and Maintenance Page: Displays the communication status, power status, and self-test results of each hardware module (such as sensors, data acquisition stations, and actuators). Provides sensor calibration entry, data storage space viewing, and log export functions.
[0123] Remote communication and maintenance gateway: In terms of hardware, this is implemented as an industrial protocol gateway integrated within a control cabinet or an edge computing device with cellular network capabilities. Its main functions include:
[0124] Transparent data transmission: The gateway communicates with the data storage management unit of the local intelligent analysis and control module via the intranet or VPN, and uploads compressed and encrypted key data packets (including status summaries, alarm records, and event reports) to a designated cloud server or enterprise-level remote monitoring center periodically (e.g., every 5 minutes) or in real time (for alarm events). The transmission protocol adopts standardized IoT protocols (such as MQTT, HTTPS) or industrial communication protocols (such as OPCUA).
[0125] Remote monitoring and configuration: Authorized users can access the cloud data platform through a secure web client or dedicated desktop application. The platform provides richer analytical tools than local HMIs, such as performance comparisons of multiple devices, long-term statistical analysis reports, and efficiency KPI dashboards. Simultaneously, engineers can remotely log in to the system (through a secure tunnel) and modify parameters remotely within their authorized scope, such as updating the weights of the fuzzy logic rule base, adjusting the sensitivity parameters of the neural network model, and optimizing action parameters (such as impact pressure and oscillation duration) in the congestion clearing strategy library.
[0126] Remote diagnostics and technical support: When the system experiences complex failures or performance degradation, remote engineers can initiate remote diagnostic sessions. This feature allows, with on-site permission, remote viewing of detailed system data streams in near real-time, downloading historical log files, and even running specific diagnostic scripts online. Combined with a rich library of historical data and case studies stored in the cloud, this aids in root cause analysis. Furthermore, the system supports remote firmware upgrades (FOTA), allowing software update packages to be securely distributed and installed.
[0127] Security and Access Control: Remote communication employs encryption (such as TLS / SSL) and authentication mechanisms throughout the entire process. The cloud platform and remote access functions implement strict role-based access control (RBAC), differentiating between different roles such as operators, maintenance engineers, and system administrators to ensure data security and operational compliance.
[0128] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
Claims
1. A method for preventing blockage in a sampling machine, characterized in that, The method includes: S1: Real-time monitoring and data acquisition. A heterogeneous multi-sensor array is deployed in the key parts of the sampler that are prone to clogging to collect multi-source operating status data, including material flow rate, equipment vibration, motor current, internal pressure and images, and transmit the collected data to the central processing unit. S2: Intelligent analysis and risk assessment of blockage characteristics. It receives multi-source operating status data collected by S1, analyzes it through a pre-trained multimodal data fusion diagnostic model, and outputs the blockage confidence, risk level and blockage type judgment for each monitoring point. S3: Adaptive congestion clearing strategy decision and execution. Based on the congestion confidence and risk level output by S2, it calls the preset strategy decision tree to generate and execute the corresponding congestion clearing control strategy. S4: Effectiveness evaluation and strategy optimization. After the congestion clearing action is executed, key parameters are collected and compared with the state before the congestion clearing, the congestion clearing efficiency index is calculated, and the data of this congestion event is stored in the historical database for the purpose of optimizing the diagnostic model and congestion clearing strategy.
2. The anti-blocking control method for the sampling machine according to claim 1, characterized in that, In S2, the multimodal data fusion diagnostic model adopts a hybrid architecture that combines a fuzzy logic preprocessing unit with a deep neural network classification unit. The fuzzy logic preprocessing unit performs fuzzification and rule reasoning on the feature quantities from S1 based on a preset expert rule base, and outputs a fuzzy feature vector representing the tendency of the blocking type. The deep neural network classification unit receives and fuses the fuzzy feature vector and the preprocessed original feature vector, performs analysis and processing, and outputs the final blockage confidence, risk level, and probability distribution of blockage type.
3. The anti-blocking control method for the sampling machine according to claim 1, characterized in that, The S3 step includes: when the blockage confidence level is lower than the first threshold, the system operates in prevention mode, and only performs data monitoring and trend analysis; When the congestion confidence level reaches or exceeds the first threshold but is below the second threshold, the system enters the early warning intervention mode and performs non-destructive intervention actions. When the congestion confidence level reaches or exceeds the second threshold, the system enters the active congestion clearing mode and performs the corresponding physical congestion clearing action according to the congestion type determined by S2.
4. The anti-blocking control method for the sampling machine according to claim 3, characterized in that, In the active unblocking mode, at least one of the following actions is performed based on the type of blockage: For adhesive blockages, activate a high-frequency micro-amplitude pneumatic oscillator; For bridging blockages, activate the directional low-frequency high-energy pneumatic impactor; For blockages caused by debris, control the micro-mechanical probe mechanism to probe or move it.
5. The anti-blocking control method for the sampling machine according to claim 1, characterized in that, In S4, the system runs a closed-loop learning and optimization engine, periodically using event records in the historical database to incrementally learn or retrain the diagnostic model in S2, and statistically analyzes and optimizes the congestion clearing strategy parameters in S3 to improve congestion clearing efficiency.
6. An intelligent anti-blocking system for a sampling machine, characterized in that, include: The distributed monitoring module is deployed in the key, easily clogged parts of the sampler to collect multi-source operational status data; The intelligent analysis and control module is connected to the distributed monitoring module, receives and analyzes the operating status data, outputs the blockage confidence level, risk level and blockage type judgment, and generates blockage clearing strategy instructions; The multimodal execution unblocking module is connected to the intelligent analysis and control module, receives and executes the unblocking strategy instructions, and performs corresponding physical unblocking actions. The intelligent analysis and control module also includes a data storage and management unit, which stores real-time operating data, historical event records and model parameters, and supports a closed-loop learning and optimization engine to optimize the diagnostic model and congestion clearing strategy.
7. The intelligent anti-clogging system for the sampler according to claim 6, characterized in that, The distributed monitoring module includes: a material flow rate monitoring unit, a mechanical load and status monitoring unit, an internal pressure monitoring unit, a visual status monitoring unit, and an environmental condition monitoring unit; each unit is connected to the intelligent analysis and control module via a real-time industrial Ethernet fieldbus.
8. The intelligent anti-clogging system for the sampling machine according to claim 6, characterized in that, The intelligent analysis and control module is deployed in an industrial-grade edge computing controller. Its software architecture includes: a data access and preprocessing unit, which receives and preprocesses data from the distributed monitoring module; The multimodal congestion diagnosis algorithm unit uses a fuzzy neural network hybrid model for real-time diagnosis; The adaptive control strategy generation unit calls the strategy library and generates clearing control commands based on the diagnostic results.
9. The intelligent anti-blocking system for the sampler according to claim 6, characterized in that, The multimodal unblocking module includes at least one of the following: a micro-pulse airflow generator array, a high-frequency micro-amplitude aerodynamic oscillator array, a directional low-frequency high-energy aerodynamic impactor, and a micro-mechanical probe mechanism.
10. The intelligent anti-blocking system for the sampling machine according to claim 6, characterized in that, Also includes: The sampling process coordinator communicates and interlocks with the main control system of the sampling machine during the clearing action to coordinate the clearing action and the sampling process. The human-computer interaction and remote operation and maintenance module provides local status display, alarm, and manual operation interface, and supports remote data transmission, configuration and diagnostic functions.