A 3D printing equipment remote cloud monitoring and intelligent operation and maintenance method and system

Abstract

The application relates to the technical field of intelligent monitoring and operation and maintenance, and discloses a 3D printing equipment remote cloud monitoring and intelligent operation and maintenance method and system, which collects physical characteristics of electromechanical units, three-dimensional coordinates of instruction codes and slice semantic attributes, synchronously splices and constructs a dynamic space-time collaborative tensor according to a global clock timestamp, calls an adaptive threshold value matched with the semantic attributes, simultaneously solves a thermal engine coupling index and a vibration integral to execute double-dimension abnormality judgment, calculates a parameter compensation increment when the index exceeds a boundary, executes safety limiting clamping in combination with a limit carbonization temperature and a minimum feeding rate, injects a correction instruction after limiting, renders a cloud twin thermal map layer through matrix transformation of the space coordinates and the index, extracts local network prediction gradients and uploads the local network prediction gradients after encryption, and finally aggregates and issues an evolution model through federal averaging. In the application, the problem of cross-source data time sequence deviation is solved, and a regulation and control closed loop of entity boundary constraint and a privacy protection evolution channel are established.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring and operation and maintenance technology, specifically to a method and system for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment. Background Technology

[0002] The operation of additive manufacturing equipment involves control code parsing and electromechanical unit linkage execution. Current monitoring schemes collect surface image characterization and basic heating parameters separately. When introducing electromagnetic torque, mechanical wave vibration signals, and 3D slice code data streams into equipment monitoring, a frequency barrier exists between the continuously sampled physical feedback electrical signals and the discretely triggered control statement sequences. The lack of a unified time reference mechanism for sequence samples obtained from different sources leads to timestamp deviations. Furthermore, the fluctuations in underlying electromechanical characteristics deviate from spatial coordinate constraints, hindering the analysis process from establishing a mapping relationship between physical state and geometric position.

[0003] The evaluation process for operational deviations uses a unified fixed baseline. The sliced ​​structure includes an outer contour entity, internal mesh filling, and a support frame. The basic physical resistance of the end effector differs during movement in different structural regions. A single evaluation condition can lead to false triggering or missed alarm intervention actions. The closed-loop intervention process after identifying operational deviations directly injects compensation and correction parameters into the controller firmware. It lacks a condition interception link based on the corresponding entity's tolerance limit. Control commands exceeding the upper limit of the polymer material's phase transition temperature range or falling below the stepper motor's holding torque limit can cause material carbonization blockage and electromechanical step loss failures. The distributed manufacturing node cluster continuously accumulates entity operation samples. Building an evolution state prediction model requires the aggregation of cross-regional node feature parameter sets. The centralized uploading of original physical acquisition sequences and code texts touches on the privacy of processing technology features, cutting off the evolution channel of the multi-node collaborative model. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment. The problem it solves is to eliminate the timing deviation between continuous physical features and discrete control commands, establish an adaptive closed-loop mechanism that adapts to slice semantics and is controlled by the physical limit boundary of the entity, and connect the model evolution and three-dimensional state space mapping channels that take into account data privacy.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of this invention provides a method for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment, applied in an architecture including edge computing nodes and a cloud monitoring platform, specifically including the following steps: The edge computing node collects physical feature data of the electromechanical actuators of the 3D printing equipment in real time. The physical feature data includes at least the electromagnetic features of the extruder stepper motor, the structural acoustic features, and the control signal sequence of the hot end heater. At the same time, it parses the printing control commands to extract three-dimensional spatial coordinates and slice semantic attributes, and synchronizes the three-dimensional spatial coordinates, the slice semantic attributes, and the physical feature data with a unified time reference to construct a dynamic spatiotemporal cooperative tensor.

[0006] The edge computing node calculates the thermomechanical coupling index based on the dynamic spatiotemporal co-current tensor and uses the corresponding adaptive threshold to evaluate abnormal states according to the slice semantic attributes. When the abnormal triggering condition is met, an abnormal flag is generated.

[0007] In response to the generation of the anomaly flag, the edge computing node initiates a closed-loop compensation mechanism to adjust the control parameters of the 3D printing equipment, and uploads the three-dimensional spatial coordinates and the thermomechanical coupling index at the moment the anomaly flag is generated to the cloud monitoring platform, so as to map them onto the three-dimensional digital model in the cloud monitoring platform for three-dimensional digital twin mapping.

[0008] Furthermore, to achieve physical quantification, the edge computing node extracts current-torque ripple components within a preset frequency range as electromagnetic features from the extruder stepper motor drive circuit, extracts mechanical friction and vibration signals from the equipment frame as structural acoustic features, and extracts the dynamic duty cycle sequence of the pulse width modulation signal as the control signal sequence. By setting a sliding time window, the variance of the current-torque ripple components and the average value of the dynamic duty cycle sequence within the window are extracted to calculate the thermomechanical coupling index. The specific calculation formula is as follows: ; In the formula, Characterizing the thermo-mechanical coupling index; This is the variance of the current torque ripple components extracted within the current sliding time window, used to reflect the motor load fluctuations caused by the melt rheological resistance inside the extruder. This represents the average value of the PWM dynamic duty cycle sequence of the hot-end heater within the same time window. and These are the normalized weighting coefficients. The above... and The specific values ​​were obtained by multiple linear regression calibration based on historical printing experience datasets, which is used to eliminate numerical differences caused by different sensor dimensions.

[0009] Furthermore, to address the false alarm problem caused by nonlinear fluctuations in physical characteristics across different printed areas of complex models (such as the shell, filling, and support), this invention introduces an adaptive threshold mechanism. The adaptive threshold includes coupling tolerance and an acoustic integral benchmark. The edge computing node dynamically calls the corresponding threshold for comparison based on the category of the current slice's semantic attributes. The mathematical logic of the anomaly triggering condition is shown in the following equation: ; In the formula, Represents the semantic category of the current slice The corresponding coupling tolerance; Characterizing mechanical friction and vibration signals at time 10:00 The amplitude; Characterizes the set sliding time window length; Characterized in length of The integral value of mechanical friction and vibration signal within the sliding time window; Represents the semantic category of the current slice The corresponding acoustic integral benchmark. The above. and The parameter mapping table is obtained by extracting boundary values ​​based on semantic classification from historical time-series data of a large number of normal printouts. An anomaly triggering condition is determined to be met only if both of the above inequalities are true simultaneously, thereby significantly reducing false alarms caused by single-dimensional feature mutations.

[0010] In terms of closed-loop control and digital twins, after an anomaly flag is generated, the edge computing node... Deviation The difference is used to calculate the temperature compensation increment and the acceleration attenuation coefficient, and correction instructions are injected into the control firmware until... Falling back to no higher than the corresponding category Within the specified range, the cloud-based monitoring platform maps the received three-dimensional spatial coordinates and corresponding thermomechanical coupling indices onto the surface of the three-dimensional digital model, and generates a thermodynamic color layer based on the spatial distribution of the values, thereby visualizing the internal rheological state and physical load.

[0011] Furthermore, in the edge-cloud collaborative evolution architecture, the edge computing nodes use their accumulated dynamic spatiotemporal collaborative tensors as samples to train the node-side decay prediction model, and after completing the training communication rounds, only the model's weight gradient updates are extracted and encrypted before being uploaded. The cloud monitoring platform executes a federated average aggregation algorithm on the multiple connected edge computing nodes. The aggregation rules for the global predictive maintenance model parameter weights are as follows: ; In the formula, The updated cloud-based global model parameter weights; These are the weights of the global model parameters from the previous round; The total number of edge computing nodes participating in this round of communication; For the first The amount of sample data used for local training by each edge computing node; The sum of the sample data volume of all participating nodes; For the first The weight gradient updates uploaded by each edge computing node. Through this operation, the data volume is used as a weight factor for aggregation, and then the updated model is distributed, achieving collective evolution of the model while protecting data privacy.

[0012] A second aspect of the present invention provides a remote cloud monitoring and intelligent operation and maintenance system for 3D printing equipment, used to implement the method described in the first aspect above. The system includes edge computing nodes and a cloud monitoring platform that are interconnected.

[0013] The edge computing node includes a first processor and a first memory. The first memory stores a first computer program executed by the first processor. When the first processor executes the first computer program, it implements the method steps of data acquisition, tensor construction, anomaly detection and evaluation, and closed-loop compensation control performed by the edge computing node.

[0014] The cloud monitoring platform includes a second processor and a second memory. The second memory stores a second computer program executed by the second processor. When the second processor executes the second computer program, it implements the method steps of three-dimensional digital twin mapping and federated average aggregation executed by the cloud monitoring platform.

[0015] This invention provides a method and system for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment. It has the following beneficial effects: 1. This invention uses a global clock to synchronously splice continuously sampled electromagnetic acoustic features with discretely extracted instruction coordinates and slice semantic attributes to generate a dynamic spatiotemporal cooperative tensor. This establishes a data mapping relationship that spans software instruction logic and electromechanical unit bottom-level feedback, eliminating timing errors caused by cross-domain frequency differences.

[0016] 2. Based on the semantic attributes of the slice, this invention calls the matching coupling tolerance and acoustic integral benchmark, and performs a two-dimensional anomaly judgment by combining the thermomechanical coupling index and vibration integral results. It associates spatial geometric features with the underlying physical tolerance and establishes a mutually restrictive trigger threshold between thermodynamic deviation and micro-friction accumulation to prevent false alarms.

[0017] 3. This invention calculates the compensation increment based on the out-of-bounds difference, and injects a correction command after executing the safety limit by combining the limit carbonization temperature cutoff and the minimum feed ratio clamping condition. This constructs a feedback closed loop from parameter adaptive control to physical boundary constraints, limiting the temperature increase and speed reduction actions to operate within the phase change range and the motor anti-stall boundary.

[0018] 4. This invention combines spatial coordinates with thermodynamic indexes to perform homogeneous transformation to render twin thermal layers, and performs local network training and encrypted gradient uploads to the cloud federated average aggregation of coverage parameters, opening up a privacy evolution channel for prediction weights across nodes, and synchronously outputting a three-dimensional spatial mapping structure with internal physical state distribution. Attached Figure Description

[0019] Figure 1 This is a diagram of the overall physical topology of the system of the present invention; Figure 2 This is the main flowchart of the method of the present invention; Figure 3 This is a flowchart of the dynamic spatiotemporal cooperative tensor construction sub-process of the present invention; Figure 4 This is the logic diagram for edge-end adaptive anomaly detection and safety closed-loop control of the present invention; Figure 5 This is a schematic diagram illustrating the evolution of the cloud-based federated learning model and twin mapping of the present invention; Figure 6 This is a block diagram of the hardware system module components of the present invention. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see the appendix Figure 1 To be continued Figure 6 This invention provides a method for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment, which is applied to an architecture that includes edge computing nodes and a cloud monitoring platform.

[0022] Edge computing nodes establish a UART serial communication connection with the underlying control motherboard of the 3D printing equipment, and simultaneously connect to a current sampling circuit and acoustic sensors via an external analog input interface. The edge computing nodes establish a data transmission channel with the cloud monitoring platform through a wide area network, forming a distributed physical monitoring topology. In this architecture, the edge computing nodes perform real-time high-frequency signal processing and closed-loop control of the underlying equipment, while the cloud monitoring platform performs data aggregation and 3D digital twin mapping tasks.

[0023] The method for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment may include the following steps: The edge computing node collects physical characteristic data of the electromechanical actuators of the 3D printing equipment in real time. The physical characteristic data includes at least the electromagnetic characteristics of the extruder stepper motor, the structural acoustic characteristics, and the control signal sequence of the hot end heater. The edge computing node parses the printing control instructions to extract three-dimensional spatial coordinates and slice semantic attributes, and synchronizes the three-dimensional spatial coordinates, slice semantic attributes and physical feature data in time series with a unified time base to construct a dynamic spatiotemporal cooperative tensor. The edge computing node calculates the thermo-mechanical coupling index based on the dynamic spatiotemporal co-current tensor, and calls the corresponding adaptive threshold to evaluate the abnormal state according to the slice semantic attributes. When it is determined that the abnormal triggering condition is met, an abnormal flag is generated. In response to the generation of the anomaly flag, the edge computing node initiates a closed-loop compensation mechanism to adjust the control parameters of the 3D printing equipment, and uploads the three-dimensional spatial coordinates and the thermomechanical coupling index at the moment the anomaly flag is generated to the cloud monitoring platform, so as to map them onto the three-dimensional digital model in the cloud monitoring platform for three-dimensional digital twin mapping.

[0024] In the overall methodology described above, edge computing nodes continuously acquire heterogeneous sensor physical signals and software control commands, establishing a collaborative relationship between the physical world and control logic. Edge computing nodes use hardware clock stamps as an alignment reference to eliminate the time asynchrony between the physical sensor sampling period and the print control command issuance period.

[0025] To quantitatively characterize the degradation state of electromechanical actuators, edge computing nodes extract correlation feature data between the load state of the extruder stepper motor and the thermodynamic state of the hot-end heater based on the synchronized dynamic spatiotemporal coordination tensor. The edge computing nodes then perform statistical ratio and weighting operations on the extracted feature data to quantify the abnormal physical coupling state inside the extruder caused by consumable carbonization or poor heat dissipation, thereby calculating the thermomechanical coupling index.

[0026] After obtaining the thermomechanical coupling index, the edge computing node uses the slice semantic attributes contained in the print control instructions to perform reasonable filtering on the current physical data fluctuations. The edge computing node invokes an adaptive threshold corresponding to the current slice semantic attributes. This adaptive threshold includes coupling tolerance and acoustic integral benchmark.

[0027] Edge computing nodes perform dual-dimensional cross-validation based on the degree to which the thermomechanical coupling index deviates from the corresponding coupling tolerance, and the degree to which the integral value of the structural acoustic features deviates from the corresponding acoustic integral benchmark within the same time window. An edge computing node is only deemed to meet the anomaly triggering condition and generates an anomaly flag if it exceeds the limits in both the thermodynamic and structural mechanical dimensions. This triggers a closed-loop parameter compensation mechanism and a cloud data upload and mapping process, completing the intelligent operation and maintenance cycle of edge-cloud collaboration.

[0028] In this embodiment, the edge computing node acquires physical characteristic data of the electromechanical actuators of the 3D printing equipment in real time, specifically including parallel acquisition and data aggregation steps of multi-dimensional physical fields. The edge computing node is allocated an independent data channel to monitor the operating status of the hardware sensing unit and the software control flow.

[0029] At the electromagnetic feature extraction level, the edge computing node acquires the stator current through a current sampling circuit at the extruder stepper motor drive circuit. Changes in the melt rheological resistance inside the hot end of the 3D printing equipment cause micro-load fluctuations in the rotor of the extruder stepper motor. These load fluctuations are directly coupled to the stator coils, manifesting as amplitude modulation of the current signal within a specific frequency band. The edge computing node dynamically determines the target characteristic frequency band based on the pole pair number parameters of the extruder stepper motor and the micro-stepping fundamental frequency of the current stepper drive pulse. The edge computing node uses a bandpass filter to perform frequency domain isolation on the acquired stator current signal, filtering out the fundamental frequency and high-frequency switching noise, and extracting the current torque ripple component within the target characteristic frequency band as the electromagnetic feature.

[0030] At the structural acoustic feature extraction level, edge computing nodes collect mechanical friction and vibration signals via acoustic sensors at the rigid support surface on the back of the extruder's moving pallet and the fixed end face of the Z-axis lead screw bearing seat. During equipment operation, the friction of the lead screw rotation, the wear of the guide rail slider, and changes in belt tension all generate mechanical waves with specific frequency characteristics. The acoustic sensors capture these mechanical waves and convert them into continuous analog voltage signals. The edge computing nodes perform analog-to-digital conversion on these analog voltage signals and output digital mechanical friction and vibration signals as the structural acoustic features.

[0031] At the control signal sequence extraction level, the edge computing node listens to the control bus between the device motherboard and the hot-end heater driver module via the SPI bus, extracting the pulse width modulation signal sent from the device motherboard to the hot-end heater, with a sampling frequency set to 1kHz. The hot-end heater adjusts its heating power according to this signal to maintain the set temperature. The edge computing node calculates the proportion of the high-level duration of the pulse width modulation signal within one cycle at a fixed sampling clock, extracting a continuous dynamic duty cycle sequence as the control signal sequence.

[0032] At the level of parsing print control commands, the edge computing node reads the G-code command text from the device motherboard cache via the aforementioned UART serial communication connection, with the command reading cycle set to 10ms. The edge computing node uses regular expressions to extract displacement parameters representing the absolute positions of the X, Y, and Z axes from the currently executed print control commands to construct the three-dimensional spatial coordinates, and extracts the F parameter representing the feed rate as the target motion velocity vector.

[0033] Simultaneously, edge computing nodes identify the semantic attributes of the current slice based on the annotation codes or trajectory topology relationships in the instruction text, and map these semantic attributes to classification elements in a preset discrete feature set. These classification elements correspond to independent category identifiers such as the shell, internal filling, support structure, and bottom bonding layer. The encoding rules for the preset discrete feature set are: shell region encoded as 0, internal filling region encoded as 1, support structure region encoded as 2, and bottom bonding layer region encoded as 3. Specifically, the identification method based on trajectory topology relationships involves extracting the X and Y axis displacement coordinates from 10 consecutive G-code instructions and calculating the movement distance of a single instruction. ; Calculate the angle of change in movement direction for consecutive commands. ,like and If the number of corners is within mm, it is determined to be a corner feature; count the number of corners within a unit length (10mm). If there are a number of corners per unit length, it is determined to be the outer shell area; if the number of corners per unit length is... And the distance moved If the Z-axis coordinate continuously increases and the distance moved is mm, it is determined to be an internal filling area; If the Z-axis coordinate is mm, it is determined to be the support structure area; if the Z-axis coordinate is 50% of the set value of the first printing layer movement speed, it is determined to be the bottom adhesive layer area.

[0034] To build a unified physical computing foundation, edge computing nodes perform the step of constructing a dynamic spatiotemporal co-current tensor. Since physical feature data is high-frequency continuously sampled data, while print control commands are low-frequency discrete event-driven data, edge computing nodes establish a strict time alignment mechanism under a set hardware clock synchronization. Edge computing nodes generate discrete timestamps using the system's global clock frequency.

[0035] Edge computing nodes will bind continuously sampled current-torque ripple components, mechanical friction and vibration signals, and dynamic duty cycle sequences carrying the same discrete timestamps to discrete three-dimensional spatial coordinates, target motion velocity vectors, and slice semantic attributes belonging to classification elements. The bound heterogeneous data streams will undergo vectorized concatenation in memory space to form the dynamic spatiotemporal collaborative tensor. This tensor will be time-stamped at discrete timestamps. The mathematical expression for time is as follows: ; In the formula, Characterizing discrete timestamps The dynamic spatiotemporal co-operation tensor at any given moment; Characterizes the current torque ripple component extracted at this moment; The numerical values ​​of the mechanical friction and vibration signals extracted at that moment; The value representing the dynamic duty cycle extracted at that moment; The three-dimensional spatial coordinate vector representing the moment; The target velocity vector at that moment; This represents the classification element encoding corresponding to the semantic attributes of the slice at that moment. The values ​​of the above variables are obtained in real time by acquiring physical sensor signals or parsing underlying text instructions through edge computing nodes. This tensor serves as the standard input data format for thermo-mechanical coupling operations and adaptive anomaly assessment.

[0036] In this embodiment, the edge computing node calculates the thermomechanical coupling index based on the constructed dynamic spatiotemporal co-current tensor. The edge computing node sets a sliding time window with a fixed step size of 100ms in the time dimension, the sliding step size being consistent with the sensor sampling period, and extracts the corresponding dimension data from the dynamic spatiotemporal co-current tensor within the current sliding time window. The edge computing node calculates the variance of the current-torque ripple components within the current sliding time window, and simultaneously calculates the average value of the dynamic duty cycle sequence.

[0037] After obtaining the aforementioned basic values, the edge computing node multiplies the variance of the current-torque ripple component and the average value of the dynamic duty cycle sequence by the corresponding normalized weighting coefficients. Subsequently, the edge computing node performs a ratio calculation to obtain the thermo-mechanical coupling index. The specific calculation process of this thermo-mechanical coupling index is characterized by the following formula: ; In the formula, Characterizes the calculated thermo-mechanical coupling index; The variance of the current torque ripple component within the current sliding time window; The average value of the dynamic duty cycle sequence within the same sliding time window; The normalized weighting coefficients that characterize the variance values; The normalized weighting coefficients represent the average value. (The above...) and The specific values ​​are based on the historical experience dataset of similar equipment during the fault-free operation phase extracted in the early stage. The measured standard calibration thrust baseline is used as the regression target dependent variable, and the variance of the current torque ripple component and the average value of the dynamic duty cycle sequence within the corresponding time period are used as the regression independent variables. The values ​​are obtained by fitting and calibrating using a multiple linear regression algorithm, which plays a role in eliminating the differences in the dimensions of data from various heterogeneous sensors.

[0038] After completing the numerical calculation, the edge computing node uses the corresponding adaptive threshold to evaluate abnormal states based on the slice semantic attributes. The adaptive threshold includes coupling tolerance and an acoustic integral benchmark. The edge computing node reads the classification element identifiers of the slice semantic attributes recorded in the current dynamic spatiotemporal co-equal tensor. Based on the category of the currently extracted slice semantic attributes, the edge computing node performs a matching query in a locally stored threshold mapping table, dynamically adjusting the allowable fluctuations of the coupling tolerance and the acoustic integral benchmark.

[0039] The printing commands for the outer shell boundary and the internal filling of the 3D model differ in feed rate and filament extrusion rate. These differences result in different distributions of the inherent rheological resistance baseline and mechanical vibration baseline under different slice categories. Edge computing nodes, by executing the logic described above—which dynamically adjusts thresholds based on slice semantic attribute categories—filter out feature value fluctuations caused by regular path switching, establishing an objective anomaly detection baseline.

[0040] After adjusting the threshold parameters, the edge computing node performs anomaly trigger verification. The edge computing node determines whether the thermomechanical coupling index exceeds the coupling tolerance of the current category, and simultaneously determines whether the integral value of the mechanical friction and vibration signals within the same sliding time window is greater than the acoustic integral reference. The mathematical logic for the above anomaly triggering conditions is represented by the following simultaneous inequalities: ; In the formula, Represents the semantic category of the current slice The corresponding coupling tolerance; Characterizing the time within the same sliding time window The amplitude values ​​of the mechanical friction and vibration signals; Characterizes the step size of the sliding time window; The integral value representing the mechanical friction and vibration signals within the same sliding time window; Represents the semantic category of the current slice The corresponding acoustic integral reference. The above parameters. and The specific values ​​are based on the historical time-series feature data without anomalies under different slice categories. The Raida statistical criterion is used to calculate the historical statistical average and standard deviation of the feature values ​​under each slice category. The historical statistical average plus three times the standard deviation is used as the upper limit statistical boundary to generate and store the threshold parameter.

[0041] When both conditions in the above simultaneous inequalities are met—that is, when the edge computing node determines that the thermomechanical coupling index exceeds the coupling tolerance of the current category, and the integral value of the mechanical friction and vibration signals within the same sliding time window is greater than the acoustic integral reference—the edge computing node determines that the abnormal triggering condition is met. After determining that the abnormal triggering condition is met, the edge computing node performs a state update operation, assigning the abnormal flag bit in the internal register to a valid state, thus completing the intelligent evaluation process at the edge.

[0042] In this embodiment, in response to the generation of the anomaly flag, the edge computing node initiates a closed-loop compensation mechanism to adjust the control parameters of the 3D printing equipment. The edge computing node extracts the thermomechanical coupling index calculated at the current moment and extracts the coupling tolerance of the corresponding slice semantic category. During the printing process, the edge computing node calculates the temperature compensation increment and the acceleration attenuation coefficient based on the difference between the thermomechanical coupling index and the adaptive threshold.

[0043] The edge computing node's built-in proportional adjustment logic executes the above calculation steps. The specific calculation logic for the temperature compensation increment and acceleration attenuation coefficient is represented by the following formula: ; ; In the formula, Characterizes the calculated temperature compensation increment; The attenuation coefficient characterizing the calculated acceleration; Characterizes the currently extracted thermo-mechanical coupling index; The coupling tolerance in the adaptive threshold corresponding to the semantic category of the current slice; Characterizes the temperature regulation proportional coefficient; Characterizes the acceleration adjustment proportional coefficient. The above parameters... and The values ​​are set and stored in a local register based on historical control test data from the earlier system integration and commissioning phase. These registers are used to convert the dimensionless exponential difference into corresponding physical control parameters. The general value range is 0.5–3℃ / unit index. The general value range is 0.1 to 0.6 per unit exponent, where the unit exponent is the thermo-mechanical coupling exponent. The value increases by 1 dimensionless unit.

[0044] After completing the control parameter calculation, the edge computing node injects a correction command into the control firmware. To prevent hardware damage caused by extreme values, the edge computing node executes hardware safety limiting judgment logic before injection. When the calculated total target temperature after merging the compensation increments exceeds the current consumable's set limit carbonization temperature, the edge computing node forcibly truncates the temperature command word in the command data packet to this limit carbonization temperature; when the calculated attenuation coefficient is lower than the preset minimum feed rate, it is clamped to this minimum feed rate. The limit carbonization temperature has a general range of 230–270°C, and the minimum feed rate has a general range of 0.25–0.3. The edge computing node sends a command data packet carrying the aforementioned safety-limited increment and coefficient control word to the underlying control motherboard via the serial communication bus. The underlying control firmware executes the correction command, using the compensation increments to increase the target temperature of the hot-end heater, thereby increasing the melt flowability of the consumable within the hot-end cavity. Simultaneously, the underlying control firmware adjusts the acceleration of the target motion velocity vector according to the attenuation coefficient to reduce the instantaneous feed thrust exerted by the extruder stepper motor on the consumable filament. The edge computing node cyclically monitors the current feature data, maintaining this closed-loop parameter adjustment state until the thermomechanical coupling index falls back to a range not exceeding the coupling tolerance of the corresponding category, and then restores the initial set control parameters.

[0045] During the local closed-loop control synchronization phase, the edge computing node uploads the three-dimensional spatial coordinates and the thermomechanical coupling index at the time the anomaly flag is generated to the cloud monitoring platform. The edge computing node packages the currently recorded three-dimensional spatial coordinate vector and corresponding numerical values ​​and transmits them via an encrypted network channel. The cloud monitoring platform receives this data packet and maps it onto the three-dimensional digital model within the cloud monitoring platform to perform a three-dimensional digital twin mapping.

[0046] The mapping to the 3D digital model within the cloud monitoring platform for 3D digital twin mapping is specifically executed by the rendering server of the cloud monitoring platform. The cloud monitoring platform pre-establishes a homogeneous coordinate transformation matrix between the origin of the actual printing physical space and the origin of the virtual digital model's bounding box. Using this spatial mapping mechanism, the cloud monitoring platform multiplies the received 3D spatial coordinates by the homogeneous coordinate transformation matrix, transforming them to the virtual spatial coordinate system to locate the mesh vertices on the surface of the cloud 3D digital model. The cloud monitoring platform assigns the corresponding thermomechanical coupling index value to each mesh vertex. Based on the spatial distribution of the thermomechanical coupling index values, the cloud monitoring platform renders a thermodynamic color layer on the surface of the 3D digital model. This layer establishes a mapping relationship between feature value magnitudes and pixel color levels, displaying the distribution of thermomechanical anomalies in different printing areas on the digital model space interface.

[0047] In this embodiment, the method further includes an edge-cloud collaborative model evolution step. The edge computing node extracts the time series of the dynamic spatiotemporal collaborative tensor accumulated in its local memory. The edge computing node uses this time series as an input sample and inputs it into the temporal neural network model deployed on the edge computing node side to perform local training calculations. The training label of the model is the predicted value of the thermomechanical coupling index 10 seconds after the end of the input time series. Supervised training is performed by using the measured thermomechanical coupling index at the corresponding time in historical data as the supervision label. The temporal neural network model specifically adopts a Long Short-Term Memory (LSTM) network architecture. The edge computing node uses the forget gate and input gate logic inside the LSTM network to calculate the historical weight memory state of the dynamic spatiotemporal collaborative tensor at different time steps to extract the long-term temporal dependency of electromechanical decay features. The edge computing node uses the gradient data of the loss function calculated from the output to update the connection parameters between the network nodes inside the local model to obtain the node-side decay prediction model. The LSTM model has an input dimension of 6, 2 hidden layers × 64 neurons, and an output dimension of 1; it uses the MSE loss function and the Adam optimizer with a learning rate of 0.001.

[0048] After the designated training communication rounds are completed, the edge computing node extracts the weight gradient update of the node-side decay prediction model. The training communication rounds are triggered every 24 hours or when the local sample size reaches 10,000. The edge computing node uses a pre-configured RSA-2048 asymmetric encryption algorithm and the public key distributed by the cloud monitoring platform to perform serialization and encryption on the extracted weight gradient update. Subsequently, the edge computing node uploads the encrypted weight gradient update to the cloud monitoring platform. The edge computing node retains the original dynamic spatiotemporal co-current tensor data in its local storage medium without transmitting it externally.

[0049] The cloud-based monitoring platform receives the weight gradient updates uploaded by multiple connected edge computing nodes at the network interface. Simultaneously, the cloud-based monitoring platform parses the data packets to obtain the actual amount of training samples used by each edge computing node during this local training process. Using the private key stored in its local security module, the cloud-based monitoring platform decrypts the received data and places the decrypted weight gradient updates into a computation register.

[0050] The cloud monitoring platform uses the amount of training samples from each edge computing node as a weighting factor to perform a federated average aggregation algorithm on all received weight gradient updates. The cloud monitoring platform weights the gradient update data uploaded by each node according to the proportion of its sample data to the total data volume, obtaining the updated parameter weights of the cloud-based global predictive maintenance model. The specific operational logic of this federated average aggregation algorithm is represented by the following formula: ; In the formula, The parameter weights characterize the updated cloud-based global predictive maintenance model; The parameter weights represent the cloud-based global predictive maintenance model retained from the previous training communication round; The total number of edge computing nodes participating in the current round of communication; Characterizing the first The amount of data for training samples on each edge computing node; The sum of the training samples of all edge computing nodes participating in the current round of aggregation represents the total amount of data, and its value satisfies the formula. ; Characterizing the first The weight gradient update amount uploaded by each edge computing node. The specific values ​​of the above parameters are obtained by the system in real time by recording the network communication status and parsing the data packet content.

[0051] After performing the above aggregation calculations, the cloud monitoring platform, based on... The model parameter file on the server side is rewritten. Subsequently, the cloud monitoring platform establishes downlink communication links to each of the edge computing nodes. The cloud monitoring platform distributes the updated global predictive maintenance model to each of the edge computing nodes, replacing the local model to improve prediction accuracy. Each edge computing node receives the updated global predictive maintenance model, overwrites the parameters of its original local time-series neural network model, and prepares to execute the monitoring and evaluation task for the next cycle.

[0052] In this embodiment, in conjunction with the aforementioned method for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment, the present invention also provides a system for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment for executing the method. This system implements all the terminal data processing and cloud aggregation steps described in the claims; the system as a whole includes physically separated but interconnected edge computing nodes and a cloud monitoring platform. The edge computing nodes and the cloud monitoring platform establish an encrypted bidirectional data transmission channel via a wide area network, jointly constituting the physical execution carrier of the present invention.

[0053] The physical hardware architecture of an edge computing node mainly includes a first processor and a first memory. The first processor and the first memory establish an instruction read and data interaction link through an internal system communication bus. The first memory is internally divided into a non-volatile storage area and a random access storage area; the first memory stores a first computer program executed by the first processor, which includes a series of compiled low-level device driver code, tensor synchronization algorithm logic, and temporal neural network model files.

[0054] When the first processor executes the first computer program residing in the first memory, it implements the high-frequency local method steps executed by the edge computing node. The first processor schedules the hardware interface channel to collect the physical characteristic data of the electromechanical actuator and parses the printing control instructions in the control bus; the first processor performs a timestamp alignment operation in the internal register space to construct a dynamic spatiotemporal cooperative tensor; subsequently, the first processor calls the arithmetic logic unit to calculate the thermomechanical coupling index based on the above tensor, and calls the corresponding adaptive threshold to execute the abnormal state verification logic according to the semantic attributes of the current slice; when the abnormal triggering condition is met and an abnormal flag bit is generated, the first processor injects compensation increment and attenuation coefficient instructions into the underlying control firmware of the 3D printing equipment, and packages the characteristic values ​​at the corresponding time and sends them to the network transmission queue.

[0055] The physical architecture of the cloud monitoring platform mainly includes a second processor and a second memory. The cloud monitoring platform adopts a distributed server cluster topology; the second processor and the second memory establish a high-speed data read / write link based on the server motherboard architecture. The second memory stores a massive database of device access files and 3D rendering engine files; it also stores a second computer program executed by the second processor.

[0056] When the second processor executes the second computer program, it implements the low-frequency global method steps executed by the cloud monitoring platform. The second processor parses the three-dimensional spatial coordinates, thermomechanical coupling index, and weight gradient update amount from each edge computing node from the network receiving port; the second processor uses the graphics processing unit to map the received three-dimensional spatial coordinates onto the surface of the internally residing three-dimensional digital model mesh, and generates a thermodynamic color layer based on the numerical gradient of the thermomechanical coupling index; at the same time, the second processor extracts the training sample data of each edge computing node, performs federated average aggregation operation on all received weight gradient updates, and sends the updated parameter weight file back to each corresponding edge computing node.

[0057] The computational task space allocation and processor execution mapping logic of this system is represented by the following mathematical set expression: ; In the formula, Characterizes the set of intelligent operation and maintenance tasks to be executed in the overall system. A method mapping operator that represents the first processor executing the first computer program; Characterizes the dynamic spatiotemporal collaborative tensor data stream continuously generated at the edge; The parameter set of the node-end decay prediction model residing in the first memory; Characterizes the method mapping operator generated by the second processor executing the second computer program; Characterizes the set of three-dimensional spatial coordinates received from the cloud for digital twin mapping; A set of thermomechanical coupling indices representing the data received from each edge node; This represents the matrix of all weight gradient updates received from the cloud. Utilizing the aforementioned physically isolated processor resource design, the system decouples high-frequency real-time closed-loop control operations from low-frequency global model aggregation operations at the hardware topology level, establishing the execution boundary for edge-cloud collaboration.

Claims

1. A method for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment, applied in an architecture including edge computing nodes and a cloud monitoring platform, characterized in that, Includes the following steps: The edge computing node collects physical characteristic data of the electromechanical actuators of the 3D printing equipment in real time. The physical characteristic data includes at least the electromagnetic characteristics of the extruder stepper motor, the structural acoustic characteristics, and the control signal sequence of the hot end heater. The edge computing node parses the printing control instructions to extract three-dimensional spatial coordinates and slice semantic attributes, and synchronizes the three-dimensional spatial coordinates, slice semantic attributes and physical feature data in time series with a unified time base to construct a dynamic spatiotemporal cooperative tensor. The edge computing node calculates the thermo-mechanical coupling index based on the dynamic spatiotemporal co-current tensor, and calls the corresponding adaptive threshold to evaluate the abnormal state according to the slice semantic attributes. When it is determined that the abnormal triggering condition is met, an abnormal flag is generated. In response to the generation of the anomaly flag, the edge computing node initiates a closed-loop compensation mechanism to adjust the control parameters of the 3D printing equipment, and uploads the three-dimensional spatial coordinates and the thermomechanical coupling index at the moment the anomaly flag is generated to the cloud monitoring platform, so as to map them onto the three-dimensional digital model in the cloud monitoring platform for three-dimensional digital twin mapping.

2. The method for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment according to claim 1, characterized in that, The edge computing node collects physical characteristic data of the electromechanical actuators of the 3D printing equipment in real time, specifically including: The stator current is collected by a current sampling circuit at the stepper motor drive circuit of the extruder, and the current torque ripple component within a preset frequency range is extracted as the electromagnetic feature. Mechanical friction and vibration signals are collected by acoustic sensors at predetermined structural locations within the equipment frame as the acoustic characteristics of the structure. The dynamic duty cycle sequence of the pulse width modulation signal sent from the device motherboard to the hot end heater is used as the control signal sequence.

3. The method for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment according to claim 2, characterized in that, The construction of the dynamic spatiotemporal cooperative tensor specifically includes: Extract the target motion velocity vector from the currently executed print control command, and map the slice semantic attributes to classification elements in a preset discrete feature set; Under the set hardware clock synchronization, the continuously sampled current torque ripple components, mechanical friction and vibration signals, and dynamic duty cycle sequence are timestamped and spliced ​​with the discrete three-dimensional spatial coordinates, the target motion velocity vector, and the slice semantic attributes belonging to the classification elements to form the dynamic spatiotemporal cooperative tensor.

4. The method for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment according to claim 3, characterized in that, The calculation of the thermo-mechanical coupling index based on the dynamic spatiotemporal co-current tensor specifically includes: Set a sliding time window, calculate the variance of the current torque ripple component within the current sliding time window, and the average value of the dynamic duty cycle sequence; The thermo-mechanical coupling index is obtained by multiplying the variance of the current torque ripple component and the average value of the dynamic duty cycle sequence by the corresponding normalized weighting coefficient.

5. A method for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment according to claim 4, characterized in that, The step of evaluating the abnormal state by calling the corresponding adaptive threshold based on the semantic attributes of the slice, and generating an abnormal flag when the abnormal triggering condition is met, specifically includes: The adaptive threshold includes coupling tolerance and acoustic integral benchmark; The edge computing node dynamically adjusts the allowable fluctuation of the coupling tolerance and the acoustic integral benchmark based on the category of the currently extracted slice semantic attributes. When the edge computing node determines that the thermomechanical coupling index exceeds the coupling tolerance of the current category, and the integral value of the mechanical friction and vibration signal within the same sliding time window is greater than the acoustic integral reference, it determines that the abnormal triggering condition is met and assigns the abnormal flag to a valid state.

6. A method for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment according to claim 5, characterized in that, The edge computing node initiates a closed-loop compensation mechanism to adjust the control parameters of the 3D printing equipment, specifically including: During the printing process, the temperature compensation increment and the acceleration attenuation coefficient are calculated based on the difference between the thermomechanical coupling index and the adaptive threshold. The edge computing node injects correction instructions into the control firmware, adjusts the target temperature of the hot-end heater using the compensation increment, and reduces the acceleration of the target motion velocity vector according to the attenuation coefficient until the thermomechanical coupling index falls back to a range not higher than the coupling tolerance of the corresponding category.

7. The method for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment according to claim 1, characterized in that, The mapping of the three-dimensional digital model onto the cloud monitoring platform to perform a three-dimensional digital twin mapping specifically includes: The cloud monitoring platform uses a spatial mapping mechanism to map the received three-dimensional spatial coordinates and the corresponding thermo-mechanical coupling index onto the surface of the three-dimensional digital model. Based on the spatial numerical distribution of the thermo-mechanical coupling index, a thermodynamic color layer is generated on the surface of the three-dimensional digital model.

8. A method for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment according to claim 1, characterized in that, The method further includes: The edge computing node uses the time series of the dynamic spatiotemporal co-current tensor it has accumulated as input samples to train a temporal neural network model deployed on the edge computing node side, thereby obtaining a node-end decay prediction model. After the training communication rounds are completed, the edge computing node extracts the weight gradient update amount of the node-side decay prediction model and encrypts and uploads the weight gradient update amount to the cloud monitoring platform.

9. A method for remote cloud monitoring and intelligent operation and maintenance of 3D printing equipment according to claim 8, characterized in that, The method further includes: The cloud monitoring platform receives the weight gradient update from multiple accessed edge computing nodes; The cloud monitoring platform uses the amount of training samples from each edge computing node as a weighting factor to perform a federated average aggregation algorithm on all received weight gradient updates to obtain the parameter weights of the updated cloud-based global predictive maintenance model. The updated global predictive maintenance model with parameter weights is then distributed to each of the edge computing nodes.

10. A remote cloud monitoring and intelligent operation and maintenance system for 3D printing equipment, characterized in that, The system is used to implement the method as described in any one of claims 1 to 9; the system includes edge computing nodes and a cloud monitoring platform that are interconnected. The edge computing node includes a first processor and a first memory. The first memory stores a first computer program executed by the first processor. When the first processor executes the first computer program, it implements the method steps executed by the edge computing node. The cloud monitoring platform includes a second processor and a second memory. The second memory stores a second computer program executed by the second processor. When the second processor executes the second computer program, it implements the method steps executed by the cloud monitoring platform.

Images