Real-time remote monitoring system for main equipment of large crankshaft automated production line

Abstract

This invention relates to the field of intelligent monitoring technology for industrial equipment, specifically a real-time remote monitoring system for the main equipment of a large-scale crankshaft automated production line. The system includes modules for data acquisition, status analysis, behavior tracking, and early warning encapsulation. The acquisition module obtains real-time sensor data on vibration waveforms, noise waveforms, temperature, and current. The status analysis module employs a dynamic field mapping framework. Through an improved fuzzy cognitive graph algorithm, it constructs dynamic concept nodes and fuzzy weighted edges based on the physical coupling relationship of data and historical operating patterns, and generates a multi-dimensional status vector through fusion reasoning. Time-varying trajectory tracking is performed on the vector to capture trajectory mutations and pattern migrations, generating hierarchical early warning signals and linking them with relevant data encapsulation. This system achieves adaptive fusion of multi-source heterogeneous data, accurately identifying sudden and gradual equipment anomalies, and improving the completeness of remote monitoring status representation and the accuracy of early warning response.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for industrial equipment, and in particular to a real-time remote monitoring system for the main equipment of a large crankshaft automated production line. Background Technology

[0002] The main equipment operation monitoring of large crankshaft automated production lines mainly uses independent sensors to collect vibration waveforms, noise waveforms, temperature signals, and current signals, and determines the equipment operating status by comparing single parameter thresholds. Multi-source sensor data processing mostly adopts fixed-structure fuzzy cognitive graphs or simple weighted fusion methods. The remote monitoring terminal only receives a single early warning command, and sensor data, status information, and early warning information are transmitted independently.

[0003] The physical coupling between sensor data is not incorporated into the analysis logic; the concept nodes and fuzzy weight edges of the fuzzy cognitive graph are fixed and cannot be adaptively adjusted based on the device's historical operating modes. Multi-source data fusion reasoning suffers from dimensional gaps, failing to form a unified vector representing the overall operating state. Traditional methods can only identify single-parameter over-threshold anomalies, lacking a trajectory analysis mechanism for multi-dimensional state vectors, and cannot capture trajectory changes and pattern transitions in the feature space. Warning signals lack response priority differentiation; warning information is not linked and encapsulated with corresponding sensor and state data, making it impossible for remote monitoring to fully acquire comprehensive information about anomaly correlations.

[0004] It is necessary to construct a fuzzy cognitive graph structure that dynamically adjusts based on physical coupling relationships and historical operating patterns, to fuse multiple types of sensor data and generate multi-dimensional state vectors. Time-varying trajectory tracking of the multi-dimensional state vectors is required to identify trajectory abrupt changes and pattern migration events, and to establish a hierarchical early warning mechanism and a multi-type data linkage encapsulation and transmission method. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a real-time remote monitoring system for the main equipment of a large-scale automated crankshaft production line.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a real-time remote monitoring system for the main equipment of a large-scale automated crankshaft production line, comprising: The data acquisition module acquires real-time sensing data of the main equipment in the large crankshaft automated production line. The real-time sensing data includes vibration waveforms, noise waveforms, temperature signals, and current signals. The state analysis module inputs the real-time sensing data into a state analysis framework based on dynamic field mapping. It uses an improved fuzzy cognitive graph algorithm to fuse and infer the real-time sensing data. The improved fuzzy cognitive graph algorithm establishes dynamic concept nodes and fuzzy weighted edges based on the physical coupling relationship and historical operation mode of the sensing data. Based on the inference results of the improved fuzzy cognitive graph algorithm, a multi-dimensional state vector describing the overall operation state of the main device is constructed. The behavior tracking module performs time-varying trajectory tracking on the multidimensional state vector, capturing trajectory abrupt changes and pattern transition events of the multidimensional state vector in the feature space; The early warning and encapsulation module generates state early warning signals with different response priorities based on the captured trajectory mutation and pattern migration events. The state early warning signals are then encapsulated with corresponding real-time sensor data segments and multi-dimensional state vector segments to form a remote monitoring data packet.

[0007] As a further aspect of the present invention, an improved fuzzy cognitive graph algorithm is used to fuse and infer real-time sensor data, including: The data acquisition module receives real-time sensing data including vibration waveform, noise waveform, temperature signal, and current signal. The real-time sensing data is normalized to obtain the normalized instantaneous value corresponding to each sensing data. The normalized instantaneous value of each sensor data is assigned to the node activation value of the corresponding vibration waveform concept node, noise waveform concept node, temperature signal concept node, and current signal concept node in the fuzzy cognitive map. Based on the predefined physical coupling relationship and historical operation mode, determine the initial strength of the fuzzy weight edge between each concept node; In the iterative reasoning process of fuzzy cognitive graph, the activation value of each concept node at the next moment is calculated based on the current activation value of each concept node and the current strength of the fuzzy weight edge. The activation value of each concept node at the next moment is the weighted sum of the current activation values ​​of all its neighboring concept nodes and the strength of the fuzzy weight edge between the concept node and its neighboring nodes, and then obtained after being mapped by a nonlinear activation function. Continue iterative reasoning until the change in activation value of all concept nodes is less than the preset convergence threshold, or the preset maximum number of iterations is reached, at which point the reasoning process terminates. From the fuzzy cognitive graph that has reached steady state, extract the final activation values ​​of vibration waveform concept nodes, noise waveform concept nodes, temperature signal concept nodes, current signal concept nodes, and all derived concept nodes; The final activation values ​​of each extracted concept node are combined in a preset order to generate a multi-dimensional state vector describing the overall operating state of the main device.

[0008] As a further aspect of the present invention, the improved fuzzy cognitive graph algorithm establishes dynamic concept nodes and fuzzy weighted edges based on the physical coupling relationship and historical operating patterns of sensor data, including: The vibration waveform, noise waveform, temperature signal, and current signal are used as four initial concept nodes. The activation value of each initial concept node is determined by the normalized instantaneous value of its corresponding sensor data. The mechanical structure and electrical principle of the main equipment of the large crankshaft automated production line are analyzed. It is determined that there is a physical coupling relationship between the vibration waveform concept node and the noise waveform concept node, and there is a physical coupling relationship between the temperature signal concept node and the current signal concept node. Based on this, fuzzy weighted edges connecting these four initial concept nodes are established. Extract sensor data patterns of the main device under different health states from historical operating data, define a derived concept node for each sensor data pattern, and determine the activation value of the derived concept node by a nonlinear combination of the activation values ​​of the initial concept node. In the fuzzy cognitive graph, fuzzy weighted edges are established from the initial concept node to the derived concept node; Based on real-time sensing data stream, the strength of all fuzzy weight edges is dynamically calculated. The strength of the fuzzy weight edge is adaptively adjusted according to the correlation between the instantaneous activation values ​​of the two concept nodes it connects to. All concept nodes and all dynamically adjusted fuzzy weight edges in the fuzzy cognitive graph together constitute a dynamic reasoning network for describing the causal relationships of device states.

[0009] As a further aspect of the present invention, time-varying trajectory tracking is performed on the multidimensional state vector to capture trajectory abrupt changes and mode transition events of the multidimensional state vector in the feature space, including: In the feature space, the historical positions of the multidimensional state vectors are recorded at a preset sampling frequency to form a sequence of state trajectory points; Calculate the vector between adjacent points in the state trajectory point sequence, and use it as the instantaneous motion vector of the state trajectory; Monitor the magnitude and direction changes of the instantaneous motion vector. When the magnitude of the instantaneous motion vector exceeds the acceleration threshold, or the direction change of the instantaneous motion vector exceeds the direction change threshold, determine that a trajectory change event has occurred, and record the time of the trajectory change event as the change timestamp. In the feature space, density clustering is performed on the sequence of state trajectory points to identify high-density clustering regions of state trajectory points. Each high-density clustering region is defined as a running mode. When a new multidimensional state vector continuously deviates from the high-density clustering region corresponding to the current operating mode and enters the high-density clustering region corresponding to another operating mode, a mode migration event is determined to have occurred, and the start timestamp of the migration and the target mode identifier are recorded.

[0010] As a further aspect of the present invention, the step of generating state warning signals with different response priorities based on the captured trajectory abrupt changes and pattern transition events includes: Warning levels are assigned to trajectory change events, and the warning levels are divided according to the proportion by which the magnitude of the instantaneous motion vector in the trajectory change event exceeds the acceleration threshold; Assign warning levels to mode migration events, and classify warning levels based on the correlation strength between the source operating mode and the target operating mode in historical fault data during the mode migration event; The warning level of the trajectory change event, the corresponding change timestamp, and the instantaneous motion vector data that triggered the trajectory change event are encapsulated into a first-class warning signal; The warning level of the mode migration event, the corresponding migration timestamp, the source operating mode identifier, and the target operating mode identifier are encapsulated into a second type of warning signal; The first type of warning signal and the second type of warning signal are collectively referred to as status warning signals, and their response priorities are marked according to their warning levels.

[0011] As a further aspect of the present invention, the status warning signal is encapsulated with the corresponding real-time sensor data segment and multi-dimensional status vector segment to form a remote monitoring data packet, including: Based on the timestamp of the status warning signal, a segment of real-time sensor data of a preset time length is extracted before and after the signal to form a real-time sensor data segment. Based on the timestamp of the state warning signal, a multi-dimensional state vector sequence of a preset time length is extracted forward and backward to form a multi-dimensional state vector segment. The metadata of the status warning signal, real-time sensor data segments, multi-dimensional status vector segments, and the corresponding master device's identity and data generation timestamp are combined according to a preset binary encapsulation format to generate a remote monitoring data packet.

[0012] As a further aspect of the present invention, the system further includes: The data transmission module transmits the remote monitoring data packets to the remote monitoring center in real time via an encrypted remote communication link; The data parsing module, located in the remote monitoring center, parses the received remote monitoring data packets and extracts the status warning signals, real-time sensor data fragments, and multi-dimensional status vector fragments. The simulation and display module uses the parsed multi-dimensional state vector fragments to drive a three-dimensional virtual device model to perform synchronous state simulation and visualization rendering. The parsed state warning signals, real-time sensor data fragments, and real-time rendering screen of the three-dimensional virtual device model are integrated and displayed on the visualization interface of the remote monitoring center. The real-time transmission of the remote monitoring data packets to the remote monitoring center via an encrypted remote communication link specifically includes: The remote monitoring data packets are encrypted using an asymmetric encryption algorithm to generate an encrypted data payload. A transmission frame header consisting of the master device identifier, data sequence number, and data checksum is appended to the encrypted data payload; The encrypted data payload with the transmission frame header is sent to the network port designated by the remote monitoring center through a physical network based on the Industrial Ethernet protocol.

[0013] As a further aspect of the present invention, the step of parsing the received remote monitoring data packets at the remote monitoring center includes: The remote monitoring center reads the transmission frame header and encrypted data payload from the designated network port; The integrity of data transmission is verified based on the data checksum in the transmission frame header; Using a decryption key paired with an asymmetric encryption algorithm, the encrypted data payload is decrypted to recover the original remote monitoring data packet. Following the reverse process of the preset binary encapsulation format, the metadata of the status warning signal, real-time sensor data fragments, multi-dimensional status vector fragments, master device identification and data generation timestamp are separated from the remote monitoring data packet.

[0014] As a further aspect of the present invention, the step of using the parsed multidimensional state vector fragments to drive a three-dimensional virtual device model for synchronous state simulation and visualization rendering includes: Based on the parsed main device identity, the corresponding three-dimensional virtual device model is loaded. The three-dimensional virtual device model includes the three-dimensional geometric and kinematic models of the crankshaft machining center, milling and turning machine tool, and heat treatment equipment. The numerical values ​​in the parsed multidimensional state vector fragments are mapped to the displacement, velocity, and angle parameters of each moving part in the three-dimensional virtual device model, as well as the temperature display parameters of each heating part. Based on the mapped parameters, the posture, motion animation, and color status of the 3D virtual device model are updated in real time in the 3D graphics engine to complete state simulation and visualization rendering.

[0015] As a further aspect of the present invention, the step of fusing and displaying the parsed status warning signal, real-time sensor data fragments, and real-time rendered images of the three-dimensional virtual device model on the visualization interface of the remote monitoring center includes: The visualization interface is divided into a 3D view area for the main device, a sensor data waveform area, and a warning information list area. The rendering of the 3D virtual device model is displayed in real time in the main device's 3D view area; In the sensor data waveform area, the waveform curves of vibration, noise, temperature and current in the real-time sensor data segments are plotted synchronously. In the warning information list area, the parsed status warning signals are displayed in a scrolling table format. Each row of the table contains warning time, warning level, event type and response priority information. When a user clicks on a row in the warning information list area, the time axis of the main device's 3D view area and the sensor data waveform area simultaneously jumps to the timestamp corresponding to the status warning signal, and highlights the device status and sensor data at the time corresponding to the selected warning signal.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: The state analysis framework based on dynamic field mapping can match the physical coupling and correlation of multiple types of sensor data. An improved fuzzy cognitive graph algorithm establishes dynamic concept nodes and fuzzy weighted edges according to the equipment's historical operating patterns. Vibration waveforms, noise waveforms, temperature signals, and current signals complete cross-dimensional data fusion and state inference. Dynamic nodes and weights can adaptively adjust according to operating conditions, and the inference process closely matches the actual physical operating mechanism of the equipment. The resulting multi-dimensional state vector can completely cover the multi-dimensional operating attributes of the main equipment, avoiding the data correlation fragmentation problem caused by fixed structure analysis. The state representation maintains a high degree of consistency with the actual operating state.

[0017] Multidimensional state vectors perform continuous time-varying trajectory tracking in the feature space, accurately capturing two types of abnormal events: trajectory abrupt changes and pattern transitions. Early warning signals with different response priorities are generated based on event type. These early warning signals are integrated with real-time sensor data segments and multidimensional state vector segments, transforming anomaly identification from single-parameter threshold judgment to overall state trajectory judgment, capturing both latent gradual transitions and sudden abrupt changes. Tiered early warning systems match the differentiated handling needs of remote terminals. The encapsulation structure enables synchronous transmission of early warning information and related data, allowing remote monitoring to directly obtain the original data corresponding to the anomaly and information on the entire process of state changes. Attached Figure Description

[0018] Figure 1 This is a timing diagram of the real-time status remote monitoring system for the main equipment of the large crankshaft automated production line described in this invention. Figure 2 A flowchart for multidimensional state vector time-varying trajectory tracking and event capture; Figure 3 A flowchart for generating and prioritizing status warning signals. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 This invention provides a real-time remote monitoring system for the main equipment of a large-scale automated crankshaft production line. The specific system includes: The data acquisition module is deployed on the production site. It periodically samples and acquires four types of real-time sensor data—vibration waveforms, noise waveforms, temperature signals, and current signals—using vibration sensors, acoustic sensors, temperature sensors, and current transformers installed on key parts of the main equipment. The sampling frequency is configured from 1kHz to 10kHz based on the equipment's process characteristics. The state analysis module incorporates a state analysis framework based on dynamic field mapping. This framework uses an improved fuzzy cognitive graph algorithm as its core inference engine, mapping normalized multi-source sensor data into concept node activation values. It then iteratively infers using dynamic weighted edges established by physical coupling relationships and historical operating patterns, ultimately outputting a multi-dimensional state vector representing the overall operating state of the equipment. The behavior tracking module continuously tracks the temporal sequence of the multi-dimensional state vector, calculating the instantaneous motion vector of the state trajectory in the feature space. By setting acceleration and direction change thresholds, it identifies trajectory change events and uses a density clustering algorithm to divide operating mode regions to detect mode migration events. The early warning and encapsulation module classifies early warning levels according to event type and severity, extracts associated sensor data segments and status vector segments based on event timestamps, and integrates device identification and timestamp information to encapsulate standardized remote monitoring data packets.

[0022] In one embodiment of the present invention, real-time sensor data streams of vibration waveform, noise waveform, temperature signal, and current signal are received from the data acquisition module. Each type of sensor data is processed using a maximum-minimum normalization method, linearly transforming the original values ​​to the [0,1] interval to obtain normalized instantaneous values. The normalized instantaneous values ​​of the vibration waveform, noise waveform, temperature signal, and current signal are assigned to the node activation values ​​of the vibration waveform concept nodes in the fuzzy cognitive graph. Based on the physical coupling relationship between the main equipment's mechanical transmission chain and electrical control circuit, the initial weights between vibration and noise concept nodes are pre-set to 0.6, between temperature and current concept nodes to 0.7, and between other nodes to 0.1. During iterative inference, the activation value of each concept node at the next moment is calculated as the weighted sum of the current activation values ​​of all its adjacent concept nodes and the corresponding fuzzy weight edge strength. The weighted result is mapped to the [-1,1] interval using a hyperbolic tangent activation function. The convergence threshold is set to 0.01, and the maximum number of iterations is 100. Inference stops when the change in activation value of all concept nodes is less than 0.01 in adjacent iterations or when the number of iterations reaches 100. The final activation values ​​of four initial concept nodes (vibration waveform, noise waveform, temperature signal, and current signal) and three derived concept nodes (main shaft load, thermal deformation, and wear state) are extracted from the steady-state fuzzy cognitive graph and combined into a seven-dimensional state vector in the order of [vibration, noise, temperature, current, load, thermal deformation, wear].

[0023] In practice, the data acquisition module periodically acquires real-time sensing data from the vibration sensor installed on the spindle of the large crankshaft machining center, the acoustic sensor in the cabinet, the temperature sensor of the spindle bearing housing, and the current transformer at the drive end of the spindle motor. The sampling frequency is configured to 2000Hz, generating 2000 sampling points of vibration waveform data, 2000 sampling points of noise waveform data, 10 sampling points of temperature signal data, and 10 sampling points of current signal data per second. The data acquisition module sends these real-time sensing data to the status analysis module via the fieldbus.

[0024] In practical implementation, the state analysis module receives real-time sensing data and performs normalization processing. For vibration waveform data, it extracts the instantaneous value of vibration acceleration at the current sampling moment, finds the minimum and maximum values ​​of vibration acceleration within the past 10 seconds, and uses the minimum-maximum normalization method to convert the instantaneous vibration acceleration value into a normalized instantaneous value within the range of [0,1]. For noise waveform data, it extracts the instantaneous value of sound pressure level at the current sampling moment, finds the minimum and maximum values ​​of sound pressure level within the past 10 seconds, and uses the minimum-maximum normalization method to convert the instantaneous sound pressure level value into a normalized instantaneous value within the range of [0,1]. The values ​​are converted to normalized instantaneous values ​​within the range of [0,1]. For temperature signal data, the temperature measurement value at the current sampling time is extracted, the minimum and maximum values ​​of the temperature within the past 10 seconds are found, and the minimum-maximum normalization method is used to convert the temperature measurement value to a normalized instantaneous value within the range of [0,1]. For current signal data, the effective value of the current at the current sampling time is extracted, the minimum and maximum values ​​of the effective value of the current within the past 10 seconds are found, and the minimum-maximum normalization method is used to convert the effective value of the current to a normalized instantaneous value within the range of [0,1].

[0025] In specific implementation, the fuzzy cognitive graph contains four initial concept nodes: vibration waveform concept node, noise waveform concept node, temperature signal concept node, and current signal concept node. The state analysis module assigns the normalized instantaneous value of the vibration waveform to the node activation value of the vibration waveform concept node, the normalized instantaneous value of the noise waveform to the node activation value of the noise waveform concept node, the normalized instantaneous value of the temperature signal to the node activation value of the temperature signal concept node, and the normalized instantaneous value of the current signal to the node activation value of the current signal concept node.

[0026] In practical implementation, the initial strength of the fuzzy weight edges between concept nodes in the fuzzy cognitive graph is predefined based on the physical coupling relationship of the main equipment in the large crankshaft automated production line. There is physical coupling caused by mechanical structure between the vibration waveform concept node and the noise waveform concept node, with the initial fuzzy weight edge strength set to 0.63. There is physical coupling caused by electrothermal effects between the temperature signal concept node and the current signal concept node, with the initial fuzzy weight edge strength set to 0.71. There is no direct physical coupling between the vibration waveform concept node and the temperature signal concept node, with the initial fuzzy weight edge strength set to 0.08. There is no direct physical coupling between the vibration waveform concept node and the current signal concept node, with the initial fuzzy weight edge strength set to 0.05. The initial fuzzy weight edge strength between the noise waveform concept node and the temperature signal concept node is also set to 0.05. There is no direct physical coupling between concept nodes, and the initial fuzzy weight edge strength is set to 0.07. There is no direct physical coupling between the noise waveform concept node and the current signal concept node, and the initial fuzzy weight edge strength is set to 0.06. There is no direct physical coupling between the temperature signal concept node and the vibration waveform concept node, and the initial fuzzy weight edge strength is set to 0.09. There is no direct physical coupling between the temperature signal concept node and the noise waveform concept node, and the initial fuzzy weight edge strength is set to 0.11. There is no direct physical coupling between the current signal concept node and the vibration waveform concept node, and the initial fuzzy weight edge strength is set to 0.13. There is no direct physical coupling between the current signal concept node and the noise waveform concept node, and the initial fuzzy weight edge strength is set to 0.12.

[0027] In practical implementation, the iterative reasoning process of the fuzzy cognitive graph involves calculating the activation value of each concept node at the next time step. The calculation formula is as follows: in: This represents the activation value of concept node i at time t+1. This represents the hyperbolic tangent activation function. This represents the strength of the fuzzy weighted edge pointing from concept node j to concept node i. This represents the activation value of concept node j at time t. This represents the total number of concept nodes in the fuzzy cognitive graph, in this embodiment. During the calculation, for each concept node, the sum of the products of the current activation values ​​of all its neighboring concept nodes and the corresponding fuzzy weight edge strengths is collected. Then, the summation result is input into the hyperbolic tangent activation function for mapping to obtain the activation value of the concept node at the next time step.

[0028] In practical implementation, the state analysis module sets the inference convergence threshold to 0.005 and the maximum number of iterations to 80. After each iteration, it checks the change in the activation values ​​of all concept nodes. If the change in the activation values ​​of all concept nodes is less than 0.005, the inference process terminates. If the number of iterations reaches 80, the inference process is forcibly terminated regardless of the change in the activation values ​​of the concept nodes. In practical implementation, the final activation values ​​of vibration waveform concept nodes, noise waveform concept nodes, temperature signal concept nodes, and current signal concept nodes are extracted from the fuzzy cognitive graph that has reached steady state. These final activation values ​​are then arranged in the order of vibration waveform concept node, noise waveform concept node, temperature signal concept node, and current signal concept node to generate a four-dimensional state vector describing the overall operating state of the main equipment.

[0029] In one embodiment of the present invention, vibration waveform, noise waveform, temperature signal, and current signal are used as initial concept nodes. The activation value of each initial concept node is directly assigned by the normalized instantaneous value of its corresponding sensor data. Based on the mechanical transmission path of the crankshaft machining center spindle-bearing-gearbox, it is determined that the vibration waveform concept node and the noise waveform concept node have strong physical coupling, and the fuzzy weight edge strength between them is initialized to 0.65. Based on the thermo-electric coupling characteristics of the motor driver, it is determined that the temperature signal concept node and the current signal concept node have physical coupling, and the fuzzy weight edge strength between them is initialized to 0.70. Sensor data patterns of three typical health states—normal cutting, overload operation, and bearing failure—are extracted from the historical operation database. Each pattern defines a derived concept node. The activation value of the normal cutting derived node is calculated by the weighted sum of the squares of the activation values ​​of the vibration and noise nodes. The activation value of the overload operation derived node is calculated by the product of the activation values ​​of the temperature and current nodes. The activation value of the bearing failure derived node is calculated by the absolute value of the difference between the activation values ​​of the vibration node and the noise node. In a fuzzy cognitive graph, fuzzy weighted edges are established pointing from four initial concept nodes to three derived concept nodes, with an initial strength set to 0.5. During real-time operation, the Pearson correlation coefficient of the activation values ​​of adjacent concept nodes is calculated every 60 seconds. The strength of the corresponding fuzzy weighted edge is updated using the absolute value of the correlation coefficient. All concept nodes and dynamically adjusted weighted edges constitute a causal inference network for device state. (See also...) Figure 2In the feature space, the coordinates of a multidimensional state vector are recorded every 100 milliseconds to form a sequence of state trajectory points. The magnitude of the instantaneous motion vector is obtained by dividing the Euclidean distance between adjacent points by the time interval, and the direction change value is obtained by calculating the angle between adjacent vectors. An acceleration threshold of 5 units / second² and a direction change threshold of 30° are set. When the magnitude of the instantaneous motion vector exceeds 5 or the direction change exceeds 30°, a trajectory change event is marked and a change timestamp is recorded. The DBSCAN density clustering algorithm is used on the state trajectory point sequence of the past 24 hours to identify high-density clustering regions. Each region corresponds to a stable operating mode. When a new state vector deviates from the current mode region and falls into another mode region for 10 consecutive sampling points, a mode migration event is determined, and the migration start timestamp and target mode identifier are recorded.

[0030] In practical implementation, the state analysis module establishes four initial concept nodes based on four types of sensor data: vibration waveform, noise waveform, temperature signal, and current signal. The activation value of the vibration waveform concept node is assigned after normalization of the instantaneous vibration acceleration value collected by the vibration sensor; the activation value of the noise waveform concept node is assigned after normalization of the instantaneous sound pressure level value collected by the acoustic sensor; the activation value of the temperature signal concept node is assigned after normalization of the temperature measurement value collected by the temperature sensor; and the activation value of the current signal concept node is assigned after normalization of the effective current value collected by the current transformer. In practical implementation, based on the mechanical structure of the spindle-gearbox-motor transmission chain of a large crankshaft machining center, the vibration waveform concept node and the noise waveform concept node both belong to mechanical state representation quantities and have a physical coupling relationship. The initial fuzzy weight edge strength connecting the two is 0.66. Based on the electrical characteristics of the motor stator winding heating and drive current, the temperature signal concept node and the current signal concept node have a physical coupling relationship. The initial fuzzy weight edge strength connecting the two is 0.72.

[0031] In some embodiments, sensor data patterns representing three typical health states of the main equipment are extracted from the historical operating database. The first is the normal cutting state, characterized by stable vibration amplitude, noise spectrum peaks concentrated at the fundamental frequency, slow temperature increase with operating conditions, and current fluctuations within the rated range. The second is the tool wear state, characterized by increased high-frequency vibration components, increased total sound pressure level, abnormally high spindle bearing temperature, and periodic fluctuations in the effective value of the current. The third is the early bearing damage state, characterized by increased vibration impact signals, enhanced specific harmonic components of noise, localized increase in bearing housing temperature, and increased current ripple. A derived concept node is defined for each sensor data pattern. The activation value of the normal cutting derived concept node is calculated by the weighted sum of the squares of the activation values ​​of the vibration waveform concept node and the noise waveform concept node. The activation value of the tool wear derived concept node is calculated by the product of the activation values ​​of the temperature signal concept node and the current signal concept node. The activation value of the early bearing damage derived concept node is calculated by the absolute value of the difference between the activation values ​​of the vibration waveform concept node and the noise waveform concept node. In the fuzzy cognitive graph, fuzzy weighted edges are established from four initial concept nodes to three derived concept nodes, with the initial strength uniformly set to 0.55.

[0032] In practice, the strength of the fuzzy weighted edges is dynamically adjusted based on the real-time sensor data stream. The Pearson correlation coefficient of the activation value sequences of adjacent concept nodes is calculated every 180 seconds, and the strength of the corresponding fuzzy weighted edge is updated using the absolute value of the correlation coefficient. For example, if the correlation coefficient between the activation value sequences of the vibration waveform concept node and the noise waveform concept node over the past 180 seconds is 0.68, then the strength of the fuzzy weighted edge between them is updated to 0.68; if the correlation coefficient between the activation value sequences of the temperature signal concept node and the current signal concept node over the past 180 seconds is 0.75, then the strength of the fuzzy weighted edge between them is updated to 0.75. All concept nodes and the dynamically adjusted fuzzy weighted edges together constitute a dynamic inference network describing the causal relationships of the device state.

[0033] It can be understood that the behavior tracking module records the historical position of the multidimensional state vector in the feature space. The multidimensional state vector is a state variable with seven dimensions output by the state analysis module. The seven dimensions are the final activation values ​​of the vibration waveform concept node, the noise waveform concept node, the temperature signal concept node, the current signal concept node, the normal cutting derived concept node, the tool wear derived concept node, and the early bearing damage derived concept node. The state trajectory point sequence records the coordinates of the multidimensional state vector in the seven-dimensional feature space at a sampling interval of 100 milliseconds, forming a temporally continuous state trajectory point sequence.

[0034] In practical implementation, the instantaneous motion vector of the state trajectory is calculated by dividing the coordinate difference between adjacent points in the state trajectory point sequence by the time interval. Let... Let be the coordinate vector of the state trajectory point at time t. for The coordinate vector of the state trajectory point at time t. With a sampling interval of 0.1 seconds, the instantaneous motion vector The calculation formula is: in: This represents the instantaneous motion vector at time t. express The coordinate vector of the state trajectory point at time t. This represents the coordinate vector of the state trajectory point at time t. This indicates a fixed sampling interval of 0.1 seconds. The magnitude of the instantaneous motion vector is the square root of the sum of the squares of its components, and the direction is the change in angle between the vector and the previous instant. An acceleration threshold of 4.5 units / second² and a change in direction threshold of 25° are set. When the magnitude of the instantaneous motion vector exceeds 4.5 or the change in direction exceeds 25°, a trajectory change event is identified, and the time of occurrence of the event is recorded as the change timestamp.

[0035] In some embodiments, the DBSCAN density clustering algorithm is used on the state trajectory point sequence accumulated over the past 24 hours, with a neighborhood radius ε of 1.5 units and a minimum number of points MinPts of 120. This identifies three high-density state trajectory point clusters: the first cluster corresponds to a normal stable operating mode, the second to a tool wear operating mode, and the third to a bearing abnormal operating mode. The center point coordinates of each high-density cluster represent the typical state characteristics of that operating mode. When a new multidimensional state vector leaves the high-density cluster corresponding to the current operating mode for 15 consecutive sampling points and enters the high-density cluster corresponding to another operating mode, a mode migration event is determined to have occurred, and the start timestamp of the migration and the target operating mode identifier are recorded.

[0036] In one embodiment of the present invention, see [reference] Figure 3For trajectory mutation events, a three-level warning system is implemented: Level 1 warning occurs when the instantaneous motion vector magnitude exceeds the acceleration threshold by 0%-50%; Level 2 warning occurs when it exceeds 50%-100%; and Level 3 warning occurs when it exceeds 100%. For mode migration events, a three-level warning system is also implemented: Level 1 warning occurs when migrating from a normal mode to a slightly abnormal mode; Level 2 warning occurs when migrating from a normal mode to a severely abnormal mode or vice versa; and Level 3 warning occurs when migrating from any mode to a mode with a failure probability greater than 20% in the historical fault database. The warning level, mutation timestamp, and triggering instantaneous motion vector magnitude and orientation angle data for trajectory mutation events are encapsulated into a first-type warning signal in JSON format. The warning level, migration timestamp, source operating mode identifier, and target operating mode identifier for mode migration events are encapsulated into a second-type warning signal in JSON format. Both types of warning signals are assigned a unified response priority: Level 1 warnings have a response delay of no more than 10 minutes, Level 2 warnings have a response delay of no more than 5 minutes, and Level 3 warnings respond in real time. Based on the timestamp of the status warning signal, real-time sensor data is extracted 30 seconds before and 10 seconds after, forming a 40-second data segment for vibration, noise, temperature, and current. Using the same timestamp as a reference, 50 multi-dimensional status vector sampling points are extracted before and 20 after, forming a 70-point status vector segment. The warning signal metadata, device number, data generation timestamp, sensor data segment, and status vector segment are encapsulated in TLV (Type-Length-Value) binary format. The type field identifies the data type, the length field records the number of bytes, and the value field stores the actual data content.

[0037] In practical implementation, the early warning and encapsulation module assigns early warning levels to trajectory change events. These levels are determined based on the proportion by which the magnitude of the instantaneous motion vector in the trajectory change event exceeds the acceleration threshold. The rules for classifying early warning levels for trajectory change events are shown in Table 1. Table 1: Rules for Classifying Early Warning Levels of Trajectory Abrupt Events In practical implementation, the early warning and encapsulation module assigns early warning levels to mode migration events. These levels are determined based on the correlation strength between the source and target operating modes in historical fault data. The historical fault database records the transition probabilities between different operating modes. For example, the probability of migrating from a normal operating mode to a slightly abnormal operating mode is 0.02, from a normal operating mode to a severely abnormal operating mode is 0.001, from a slightly abnormal operating mode to a severely abnormal operating mode is 0.03, and from a severely abnormal operating mode to a shutdown fault mode is 0.35. Based on the transition probabilities in the historical fault database, the early warning level classification rules for mode migration events are set as follows: a transition probability between the source and target operating modes of less than or equal to 0.025 is a Level 1 early warning; a transition probability greater than 0.025 and less than or equal to 0.035 is a Level 2 early warning; and a transition probability greater than 0.035 is a Level 3 early warning.

[0038] In practical implementation, the warning level of a trajectory mutation event, the corresponding mutation timestamp, and the instantaneous motion vector magnitude and direction angle data triggering the trajectory mutation event are encapsulated into a first type of warning signal. The first type of warning signal is organized in JSON data format and contains four fields: the "alert_type" field has the value "trajectory_sudden_change", the "level" field has the warning level string, the "timestamp" field has the mutation timestamp as a long integer value, and the "vector_data" field has a nested object containing the instantaneous motion vector magnitude and direction angle. The warning level of a pattern migration event, the corresponding migration timestamp, and the source and target operating mode identifier strings are encapsulated into a second type of warning signal. The second type of warning signal is organized in JSON data format and contains five fields: the "alert_type" field has the value "pattern_migration", the "level" field has the warning level string, the "timestamp" field has the migration timestamp as a long integer value, the "source_mode" field has the source operating mode identifier string, and the "target_mode" field has the target operating mode identifier string. The first and second types of warning signals are collectively referred to as status warning signals. Status warning signals are assigned response priorities according to the warning level. Level 1 warnings require a response within 10 minutes, Level 2 warnings require a response within 5 minutes, and Level 3 warnings require an immediate response.

[0039] In practical implementation, based on the timestamp of the state warning signal, a 30-second segment of real-time sensor data is extracted forward, and a 10-second segment is extracted backward, forming a 40-second real-time sensor data segment. This segment includes vibration waveform data, noise waveform data, temperature signal data, and current signal data. The vibration waveform data segment has a sampling frequency of 2000Hz and a total of 80,000 sampling points; the noise waveform data segment has a sampling frequency of 2000Hz and a total of 80,000 sampling points; the temperature signal data segment has a sampling frequency of 10Hz and a total of 400 sampling points; and the current signal data segment has a sampling frequency of 10Hz and a total of 400 sampling points. Based on the timestamp of the same state warning signal, 50 multi-dimensional state vector sampling points are extracted forward, and 20 multi-dimensional state vector sampling points are extracted backward, forming a 70-sampling-point multi-dimensional state vector segment. Each state vector in the multidimensional state vector segment is a seven-dimensional vector containing the final activation values ​​of the vibration waveform concept node, the noise waveform concept node, the temperature signal concept node, the current signal concept node, the normal cutting derived concept node, the tool wear derived concept node, and the early bearing damage derived concept node.

[0040] In some embodiments, the status warning signal metadata, real-time sensor data fragments, and multi-dimensional state vector fragments are combined using a TLV (Type-Length-Value) binary encapsulation format. The status warning signal metadata is encapsulated with type label 0x01, the length field being the number of bytes in the metadata JSON string, and the value field being a UTF-8 encoded byte array of the metadata JSON string. The real-time sensor data fragment is encapsulated with type label 0x02, the length field being the total number of bytes in all sensor data fragments, and the value field sequentially arranging the little-endian floating-point arrays of the vibration waveform data fragment, noise waveform data fragment, temperature signal data fragment, and current signal data fragment. The multi-dimensional state vector fragment is encapsulated with type label 0x03, the length field being the total number of bytes in all state vector data, and the value field sequentially arranging the floating-point arrays of 70 seven-dimensional state vectors, with the seven floating-point numbers of each state vector arranged in a fixed order. The master device's identity string is encapsulated as a type label 0x04, with a length field representing the number of bytes in the identity string and a value field representing a UTF-8 encoded byte array of the identity string. The data generation timestamp is encapsulated as a type label 0x05, with a length field of 8 bytes and a value field representing a 64-bit little-endian integer representing a Unix timestamp. This can be understood as the total length of the TLV encapsulation block. It can be calculated using the formula: in: This indicates the total number of bytes in the headers (type + length) of all TLV entries, with each header occupying a fixed 5 bytes; This indicates the length of the value field in the status warning signal metadata TLV entry; The value field of a real-time sensor data segment TLV entry indicates the length of the field. Indicates the length of the TLV entry value field in the multidimensional state vector fragment; Indicates the length of the value field of the master device identity TLV entry; The value field indicates the length of the data generation timestamp TLV entry. All TLV entries are sorted in ascending order by type label and combined to form a complete remote monitoring data packet.

[0041] In one embodiment of the invention, the system adds a data transmission module and a data parsing module and simulation display module for the remote monitoring center. The data transmission module uses the RSA-2048 asymmetric encryption algorithm to encrypt the entire remote monitoring data packet, generating an encrypted data payload. A 16-byte transmission frame header is appended to the payload, containing a 4-byte device identification, a 4-byte data sequence number, and an 8-byte CRC32 data checksum. The encrypted data packet is sent to the remote monitoring center's port 192.168.1.100:8080 via the Gigabit Industrial Ethernet TCP protocol. The remote monitoring center reads the transmission frame header and encrypted data payload from the port, first verifying the data integrity according to the CRC32 checksum, discarding data packets that fail the verification, and decrypting the encrypted data payload using the private key to restore the original data packet after successful verification. Reverse parsing is performed according to the TLV format: first, the type field is read to determine the data type, and the value field content of the corresponding bytes is extracted according to the length field, sequentially separating the warning signal metadata, real-time sensor data fragments, multi-dimensional state vector fragments, device identification, and data generation timestamp.

[0042] In practical implementation, the data transmission module is located inside the industrial control computer on the large crankshaft automated production line. It is responsible for sending the remote monitoring data packets generated by the early warning and packaging module outwards. The data transmission module uses the RSA-2048 asymmetric encryption algorithm, employing a pre-generated public key to encrypt the entire remote monitoring data packet. The public key is stored in the secure storage area of ​​the industrial control computer. The encryption process uses PKCS#1v1.5 padding mode, generating an encrypted data payload. In practical implementation, a transmission frame header is added to the encrypted data payload. The transmission frame header consists of three parts: a 4-byte master device identifier, using ASCII encoding to store the first 7 characters of the device number, such as "CNC_001"; a 4-byte data sequence number, an unsigned integer incremented by 1 for each data packet sent, with an initial value of 0; and an 8-byte data checksum, calculated using the CRC64-ECMA algorithm to determine the cyclic redundancy check value of the encrypted data payload. The transmission frame header and the encrypted data payload are directly concatenated to form the network transmission packet to be sent.

[0043] In some embodiments, network packets are sent via a physical network based on the Industrial Ethernet protocol. The field switch is configured with a Gigabit Ethernet interface supporting the IEEE 802.3 standard. The transmission link uses the TCP protocol to ensure reliability. The server IP address of the remote monitoring center is configured as 192.168.10.100, and the listening port is 8888. The data transmission module on the industrial control computer calls the socket interface to send the network packets to the designated port of the remote monitoring center via the TCP connection. The sending timeout is set to 5000 milliseconds, and the number of retries is set to 3.

[0044] It is understandable that the data parsing module of the remote monitoring center runs on the central server, which is bound to port 8888 and continuously listens for access requests. The data parsing module reads the incoming data stream from the network port, first extracting the first 16 bytes as the transmission frame header, and using the remaining part as the encrypted data payload. The structure of the transmission frame header is shown in Table 2. Table 2: Structure of Transmission Frame Header In practice, the data parsing module verifies data transmission integrity based on the data checksum in the transmission frame header, extracting the 8-byte CRC64 checksum field from the header, denoted as ReceivedCRC. Simultaneously, it recalculates the CRC64-ECMA checksum value for the received encrypted data payload, denoted as CalculatedCRC. If ReceivedCRC and CalculatedCRC are not equal, it indicates an error in the transmission process; the data packet is discarded and logged. If they are equal, the data integrity verification passes. After successful integrity verification, the data parsing module decrypts the encrypted data payload using a pre-stored private key. The private key is stored in the secure hardware module of the remote monitoring center. The decryption algorithm is paired with the encryption algorithm, and upon successful decryption, the original remote monitoring data packet is recovered.

[0045] In some embodiments, the decrypted remote monitoring data packets are format-parsed. The remote monitoring data packets are encapsulated using a TLV structure, and the parsing program reads the data according to the type tag order. First, type tag 0x01 is read to obtain the length field of the status warning signal metadata. Based on the length field, the value field content of the corresponding number of bytes is read, and the value field content is decoded into a JSON string to extract the detailed information of the status warning signal. Next, type tag 0x02 is read to obtain the length field of the real-time sensor data segment. Based on the length field, all sensor data bytes are read and decomposed into independent floating-point arrays in the order of vibration waveform data segment, noise waveform data segment, temperature signal data segment, and current signal data segment. Subsequently, type tag 0x03 is read to obtain the length field of the multidimensional status vector segment. Based on the length field, all status vector data is read and reassembled into a multidimensional status vector sequence in groups of 7 floating-point numbers. Finally, type tags 0x04 and 0x05 are read to obtain the master device identification string and the data generation timestamp value, respectively. All parsed data items are stored in the time-series database of the monitoring center for use by subsequent modules.

[0046] In one embodiment of the present invention, based on the parsed device identity identifier, a corresponding 3D virtual device model is loaded from the model library. The model includes a 3D CAD geometric mesh of a crankshaft machining center, a kinematic joint tree of a milling and turning machine tool, and a temperature field texture map of a heat treatment device. The vibration-derived dimension values ​​in the multi-dimensional state vector fragments are mapped to the spindle axial displacement parameters, the noise-derived dimension values ​​to the gearbox speed fluctuation parameters, the temperature-derived dimension values ​​to the motor housing temperature display values, and the current-derived dimension values ​​to the driver torque parameters. In the Unity 3D graphics engine, the translation matrix of the spindle components, the gearbox rotation angle, and the color gradient of the motor model surface (blue to red indicates temperature increase) are updated in real time according to the mapped parameters, and the driving model presents synchronized motion animation and state rendering. The visualization interface is divided into a 3D view area of ​​the main device on the left, a sensor data waveform area on the upper right, and a warning information list area on the lower right. The 3D view area refreshes the rendering screen of the 3D virtual device model at 30 frames per second, and the waveform area synchronously plots four sensor data curves: vibration, noise, temperature, and current. The horizontal axis represents the timestamp, and the vertical axis represents the normalized amplitude. The warning information list table contains four columns: warning time, warning level (Level 1 / 2 / 3), event type (trajectory change / pattern migration), and response priority. New warnings are inserted into the table header and highlighted. When a user clicks on a row in the table, the timelines in the 3D view area and waveform area automatically locate the warning timestamp, the 3D model jumps to the corresponding pose, a vertical marker line is added to the waveform area at that moment, and the component in the 3D model that triggered the event flashes as a notification.

[0047] In practice, the simulation display module loads the corresponding 3D virtual equipment model based on the main equipment identification string extracted by the data parsing module. The main equipment identification string is “LineA_Station3_CNC”. The model library stores 3D virtual equipment model files that match this identifier. The files contain 3D CAD geometric mesh data of the crankshaft machining center, kinematic joint hierarchical structure data of the turning and milling composite machine tool, and furnace surface texture coordinate data of the heat treatment equipment. In practical implementation, the values ​​in the parsed multidimensional state vector segments are mapped to the displacement, velocity, and angle parameters of each moving part in the three-dimensional virtual device model, as well as the temperature display parameters of each heating part. The multidimensional state vector segments contain seven-dimensional state vector data at 70 time points. The vibration-derived dimension values ​​in the state vectors are mapped to the axial displacement parameter of the spindle box along the Z-axis, with a value range of [-0.5mm, +0.5mm]. The noise-derived dimension values ​​are mapped to the gearbox output shaft speed fluctuation parameter, with a value range of [-15rpm, +15rpm]. The temperature-derived dimension values ​​are mapped to the surface temperature display value of the motor housing, with a value range of [25°C, 95°C]. The current-derived dimension values ​​are mapped to the servo driver output torque percentage parameter, with a value range of [0%, 150%].

[0048] In practice, the pose, motion animation, and color state of the 3D virtual device model are updated in real time in the Unity 3D graphics engine based on the mapped parameters. The displacement parameters of the spindle box are converted into translation transformation matrices in the model coordinate system, the gearbox speed fluctuation parameters are converted into the angular velocity attributes of the rotary joints, and the motor housing temperature display value is mapped to the color gradient through linear interpolation. The low temperature of 25°C corresponds to blue RGB(0,0,255), the high temperature of 95°C corresponds to red RGB(255,0,0), and intermediate temperature values ​​are mixed with color channels proportionally to complete the state simulation and visualization rendering. The visualization interface is divided into three main functional areas: the main device 3D view area, the sensor data waveform area, and the warning information list area. The main device 3D view area occupies 60% of the left side of the screen, displaying the real-time rendered image of the 3D virtual device model at a resolution of 1920×1080 pixels and a refresh rate of 30 frames per second.

[0049] In some embodiments, the sensor data waveform area is located in the upper right corner of the screen, occupying 35% of the screen height. It is divided into four sub-areas, each plotting the vibration waveform, noise waveform, temperature signal, and current signal curves from the parsed real-time sensor data segment. The horizontal axis represents the timestamp, covering a 40-second time window, and the vertical axis represents the normalized amplitude, ranging from [0,1]. The warning information list area is located in the lower right corner of the screen, occupying 25% of the screen height. It displays the parsed status warning signals in a scrolling table format. The table columns include "Warning Time," "Warning Level," "Event Type," and "Response Priority." Each row displays the occurrence time, level identifier, trajectory change or pattern migration type, and high, medium, and low priority labels of the warning event. Optionally, when a user clicks on a row in the warning information list area table, the timelines of the main device's 3D view area and the sensor data waveform area synchronously jump to the timestamp position corresponding to the selected status warning signal. The 3D virtual device model instantly updates to the attitude state corresponding to that moment, and a vertical red marker line is added to the sensor data waveform area at the corresponding time position. Simultaneously, the component in the 3D model that triggered the warning event activates a flashing highlight effect with a flashing frequency of 2Hz.

[0050] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A real-time remote monitoring system for the main equipment of a large-scale automated crankshaft production line, characterized in that, include: The data acquisition module acquires real-time sensing data of the main equipment in the large crankshaft automated production line. The real-time sensing data includes vibration waveforms, noise waveforms, temperature signals, and current signals. The state analysis module inputs the real-time sensing data into a state analysis framework based on dynamic field mapping. It uses an improved fuzzy cognitive graph algorithm to fuse and infer the real-time sensing data. The improved fuzzy cognitive graph algorithm establishes dynamic concept nodes and fuzzy weighted edges based on the physical coupling relationship and historical operation mode of the sensing data. Based on the inference results of the improved fuzzy cognitive graph algorithm, a multi-dimensional state vector describing the overall operation state of the main device is constructed. The behavior tracking module performs time-varying trajectory tracking on the multidimensional state vector, capturing trajectory abrupt changes and pattern transition events of the multidimensional state vector in the feature space; The early warning and encapsulation module generates state early warning signals with different response priorities based on the captured trajectory mutation and pattern migration events. The state early warning signals are then encapsulated with corresponding real-time sensor data segments and multi-dimensional state vector segments to form a remote monitoring data packet.

2. The real-time remote monitoring system for the main equipment of a large-scale automated crankshaft production line according to claim 1, characterized in that, An improved fuzzy cognitive graph algorithm is used to fuse and infer real-time sensor data, including: The data acquisition module receives real-time sensing data including vibration waveform, noise waveform, temperature signal, and current signal. The real-time sensing data is normalized to obtain the normalized instantaneous value corresponding to each sensing data. The normalized instantaneous value of each sensor data is assigned to the node activation value of the corresponding vibration waveform concept node, noise waveform concept node, temperature signal concept node, and current signal concept node in the fuzzy cognitive map. Based on the predefined physical coupling relationship and historical operation mode, determine the initial strength of the fuzzy weight edge between each concept node; In the iterative reasoning process of fuzzy cognitive graph, the activation value of each concept node at the next moment is calculated based on the current activation value of each concept node and the current strength of the fuzzy weight edge. The activation value of each concept node at the next moment is the weighted sum of the current activation values ​​of all its neighboring concept nodes and the strength of the fuzzy weight edge between the concept node and its neighboring nodes, and then obtained after being mapped by a nonlinear activation function. Continue iterative reasoning until the change in activation value of all concept nodes is less than the preset convergence threshold, or the preset maximum number of iterations is reached, at which point the reasoning process terminates. From the fuzzy cognitive graph that has reached steady state, extract the final activation values ​​of vibration waveform concept nodes, noise waveform concept nodes, temperature signal concept nodes, current signal concept nodes, and all derived concept nodes; The final activation values ​​of each extracted concept node are combined in a preset order to generate a multi-dimensional state vector describing the overall operating state of the main device.

3. The real-time remote monitoring system for the main equipment of a large-scale crankshaft automated production line according to claim 1, characterized in that, The improved fuzzy cognitive graph algorithm establishes dynamic concept nodes and fuzzy weighted edges based on the physical coupling relationship of sensor data and historical operating patterns, including: The vibration waveform, noise waveform, temperature signal, and current signal are used as four initial concept nodes. The activation value of each initial concept node is determined by the normalized instantaneous value of its corresponding sensor data. The mechanical structure and electrical principle of the main equipment of the large crankshaft automated production line are analyzed. It is determined that there is a physical coupling relationship between the vibration waveform concept node and the noise waveform concept node, and there is a physical coupling relationship between the temperature signal concept node and the current signal concept node. Based on this, fuzzy weighted edges connecting these four initial concept nodes are established. Extract sensor data patterns of the main device under different health states from historical operating data, define a derived concept node for each sensor data pattern, and determine the activation value of the derived concept node by a nonlinear combination of the activation values ​​of the initial concept node. In the fuzzy cognitive graph, fuzzy weighted edges are established from the initial concept node to the derived concept node; Based on real-time sensing data stream, the strength of all fuzzy weight edges is dynamically calculated. The strength of the fuzzy weight edge is adaptively adjusted according to the correlation between the instantaneous activation values ​​of the two concept nodes it connects to. All concept nodes and all dynamically adjusted fuzzy weight edges in the fuzzy cognitive graph together constitute a dynamic reasoning network for describing the causal relationships of device states.

4. The real-time remote monitoring system for the main equipment of a large-scale automated crankshaft production line according to claim 3, characterized in that, Time-varying trajectory tracking is performed on the multidimensional state vector to capture trajectory abrupt changes and mode transition events in the feature space, including: In the feature space, the historical positions of the multidimensional state vectors are recorded at a preset sampling frequency to form a sequence of state trajectory points; Calculate the vector between adjacent points in the state trajectory point sequence, and use it as the instantaneous motion vector of the state trajectory; Monitor the magnitude and direction changes of the instantaneous motion vector. When the magnitude of the instantaneous motion vector exceeds the acceleration threshold, or the direction change of the instantaneous motion vector exceeds the direction change threshold, determine that a trajectory change event has occurred, and record the time of the trajectory change event as the change timestamp. In the feature space, density clustering is performed on the sequence of state trajectory points to identify high-density clustering regions of state trajectory points. Each high-density clustering region is defined as a running mode. When a new multidimensional state vector continuously deviates from the high-density clustering region corresponding to the current operating mode and enters the high-density clustering region corresponding to another operating mode, a mode migration event is determined to have occurred, and the start timestamp of the migration and the target mode identifier are recorded.

5. The real-time remote monitoring system for the main equipment of a large-scale automated crankshaft production line according to claim 4, characterized in that, Based on the captured trajectory abrupt changes and pattern transition events, state warning signals with different response priorities are generated, including: Warning levels are assigned to trajectory change events, and the warning levels are divided according to the proportion by which the magnitude of the instantaneous motion vector in the trajectory change event exceeds the acceleration threshold; Assign warning levels to mode migration events, and classify warning levels based on the correlation strength between the source operating mode and the target operating mode in historical fault data during the mode migration event; The warning level of the trajectory change event, the corresponding change timestamp, and the instantaneous motion vector data that triggered the trajectory change event are encapsulated into a first-class warning signal; The warning level of the mode migration event, the corresponding migration timestamp, the source operating mode identifier, and the target operating mode identifier are encapsulated into a second type of warning signal; The first type of warning signal and the second type of warning signal are collectively referred to as status warning signals, and their response priorities are marked according to their warning levels.

6. The real-time remote monitoring system for the main equipment of a large-scale crankshaft automated production line according to claim 1, characterized in that, The status warning signal is encapsulated with the corresponding real-time sensor data segment and multi-dimensional status vector segment to form a remote monitoring data packet, including: Based on the timestamp of the status warning signal, a segment of real-time sensor data of a preset time length is extracted before and after the signal to form a real-time sensor data segment. Based on the timestamp of the state warning signal, a multi-dimensional state vector sequence of a preset time length is extracted forward and backward to form a multi-dimensional state vector segment. The metadata of the status warning signal, real-time sensor data segments, multi-dimensional status vector segments, and the corresponding master device identification and data generation timestamp are combined according to a preset binary encapsulation format to generate a remote monitoring data packet.

7. The real-time remote monitoring system for the main equipment of a large-scale automated crankshaft production line according to claim 1, characterized in that, The system also includes: The data transmission module transmits the remote monitoring data packets to the remote monitoring center in real time via an encrypted remote communication link; The data parsing module, located in the remote monitoring center, parses the received remote monitoring data packets and extracts the status warning signals, real-time sensor data fragments, and multi-dimensional status vector fragments. The simulation and display module uses the parsed multi-dimensional state vector fragments to drive a three-dimensional virtual device model to perform synchronous state simulation and visualization rendering. The parsed state warning signals, real-time sensor data fragments, and real-time rendering screen of the three-dimensional virtual device model are integrated and displayed on the visualization interface of the remote monitoring center. The real-time transmission of the remote monitoring data packets to the remote monitoring center via an encrypted remote communication link specifically includes: The remote monitoring data packets are encrypted using an asymmetric encryption algorithm to generate an encrypted data payload. A transmission frame header consisting of the master device identifier, data sequence number, and data checksum is appended to the encrypted data payload; The encrypted data payload with the transmission frame header is sent to the network port designated by the remote monitoring center through a physical network based on the Industrial Ethernet protocol.

8. The real-time remote monitoring system for the main equipment of a large-scale crankshaft automated production line according to claim 7, characterized in that, The process of parsing the received remote monitoring data packets at the remote monitoring center includes: The remote monitoring center reads the transmission frame header and encrypted data payload from the designated network port; The integrity of data transmission is verified based on the data checksum in the transmission frame header; Using a decryption key paired with an asymmetric encryption algorithm, the encrypted data payload is decrypted to recover the original remote monitoring data packet. Following the reverse process of the preset binary encapsulation format, the metadata of the status warning signal, real-time sensor data fragments, multi-dimensional status vector fragments, master device identification and data generation timestamp are separated from the remote monitoring data packet.

9. The real-time remote monitoring system for the main equipment of a large-scale crankshaft automated production line according to claim 8, characterized in that, The method of using the parsed multi-dimensional state vector fragments to drive a three-dimensional virtual device model for synchronous state simulation and visualization rendering includes: Based on the parsed main device identity, the corresponding three-dimensional virtual device model is loaded. The three-dimensional virtual device model includes the three-dimensional geometric and kinematic models of the crankshaft machining center, milling and turning machine tool, and heat treatment equipment. The numerical values ​​in the parsed multidimensional state vector fragments are mapped to the displacement, velocity, and angle parameters of each moving part in the three-dimensional virtual device model, as well as the temperature display parameters of each heating part. Based on the mapped parameters, the posture, motion animation, and color status of the 3D virtual device model are updated in real time in the 3D graphics engine to complete state simulation and visualization rendering.

10. The real-time status remote monitoring system for the main equipment of a large-scale crankshaft automated production line according to claim 7, characterized in that, The process of integrating and displaying the parsed status warning signals, real-time sensor data fragments, and real-time rendered images of the 3D virtual device model on the visualization interface of the remote monitoring center includes: The visualization interface is divided into a 3D view area for the main device, a sensor data waveform area, and a warning information list area. The rendering of the 3D virtual device model is displayed in real time in the main device's 3D view area; In the sensor data waveform area, the waveform curves of vibration, noise, temperature and current in the real-time sensor data segments are plotted synchronously. In the warning information list area, the parsed status warning signals are displayed in a scrolling table format. Each row of the table contains warning time, warning level, event type and response priority information. When a user clicks on a row in the warning information list area, the time axis of the main device's 3D view area and the sensor data waveform area simultaneously jumps to the timestamp corresponding to the status warning signal, and highlights the device status and sensor data at the time corresponding to the selected warning signal.

Images