Grinding robot monitoring method and system based on Internet of Things and related equipment

Abstract

The invention relates to the technical field of automation control, in particular to a grinding robot monitoring method and system based on the Internet of Things and related equipment, and the method comprises the steps that multi-element original data in the grinding process is collected in real time by deploying multiple types of Internet of Things sensors; performing cleaning, noise reduction, standardization and feature fusion on the data by using edge computing nodes to generate machine state feature vectors; inputting the feature vector into a preset dynamic adjustment model, and outputting a polishing process parameter adjustment instruction; after safety verification and amplitude limiting processing are conducted on the adjusting instruction, the adjusting instruction is issued to a bottom layer controller of the grinding robot through an industrial communication protocol to be executed; meanwhile, instruction feedback information is collected and associated with original state data, a complete monitoring data packet is constructed and uploaded to the cloud platform through the Internet of Things gateway, and visual and distributed monitoring and closed-loop management of the polishing task are achieved. According to the invention, the self-adaptive intelligent control of the polishing process can be realized, and the polishing quality and the production management efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of automation control technology, and in particular to a method, system and related equipment for monitoring grinding robots based on the Internet of Things. Background Technology

[0002] With the development of manufacturing technology, grinding robots are widely used in aerospace, automotive manufacturing, and precision parts processing. They possess advantages such as high repeatability, high consistency, and programmable control, gradually replacing traditional manual grinding operations and becoming an important means of automating surface finishing. To ensure the stability of grinding quality and the consistency of product surface treatment, process parameters during grinding (such as spindle speed, feed rate, and grinding pressure) need to be dynamically adjusted according to material properties, environmental conditions, and real-time status.

[0003] Existing grinding robot control systems still suffer from several key technical bottlenecks. On the one hand, traditional systems mostly use single or a small number of sensors for coarse-grained monitoring, lacking the fusion of multi-source high-frequency sensor data and deep feature modeling, making it difficult to accurately reflect the operating status under complex grinding conditions. On the other hand, existing parameter adjustment methods often rely on fixed rules or simple models, which cannot cope with dynamically changing grinding environments, easily leading to problems such as parameter response lag, low control accuracy, or lack of adaptive capabilities. In addition, some systems lack a closed-loop feedback mechanism for the execution process, failing to achieve full-process recording and cloud-based visual management of grinding tasks, making it difficult to identify and trace abnormal situations in a timely manner. Summary of the Invention

[0004] This invention provides a grinding robot monitoring method, system, and related equipment based on the Internet of Things (IoT). The grinding robot monitoring method integrates IoT sensing, multimodal feature extraction, intelligent decision-making models, and cloud visualization capabilities to improve the system's intelligent control accuracy and operation and maintenance management capabilities.

[0005] The IoT-based monitoring method for polishing robots includes the following steps: S1: Collect diverse raw data of the polishing process in real time through an IoT sensor network deployed on the polishing robot body and working environment; S2: The multi-source raw data is transmitted to the edge computing node for preprocessing and data fusion, and key features are extracted to generate a machine state feature vector that represents the current comprehensive state. S3: Input the machine state feature vector into a preset dynamic adjustment model for analysis and calculation, and generate grinding process parameter adjustment instructions for real-time optimization of the grinding process; S4: Send the grinding process parameter adjustment command to the bottom controller of the grinding robot to perform real-time adjustment, and at the same time upload the machine status feature vector and the grinding process parameter adjustment command to the cloud platform for centralized monitoring and recording.

[0006] Optionally, S1 includes: S11: Initialize the IoT sensor network deployed on the grinding robot body and working environment, configure the sampling frequency and data format of each sensor, and generate an initialized and ready IoT sensor network. S12: Through the initialized IoT sensor network, the initial sensor data streams of the grinding robot motor current, body vibration, grinding head acoustic emission, tool end force torque and ambient temperature and humidity are synchronously collected at a preset sampling frequency. S13: Perform real-time verification and timestamp synchronization on the initial sensing data stream to eliminate data packet loss and timing disorder, and generate synchronized and uniformly formatted multi-source raw data; S14: The synchronized and uniformly formatted multi-source raw data is framed and buffered according to a preset data packet size to generate multi-source raw data to be transmitted.

[0007] Optionally, the physical addresses and logical mapping relationships of the vibration sensors, acoustic emission sensors, force / torque sensors, and environmental temperature and humidity sensors included in the IoT sensor network are defined, and the sampling frequency and data output format of each sensor are uniformly set to complete the generation of the initialized and ready IoT sensor network.

[0008] Optionally, S2 includes: S21: The edge computing node receives the multi-source raw data to be transmitted, and performs data cleaning, noise reduction and standardization processing on the multi-source raw data to be transmitted to generate pre-processed multi-source raw data. S22: Perform spatiotemporal alignment and feature layer fusion on the preprocessed multivariate raw data to map the preprocessed multivariate raw data from different sources and types to a unified high-dimensional feature space, and generate a fused multidimensional data matrix. S23: Extract time-domain, frequency-domain, and time-frequency-domain indicators that are strongly correlated with equipment status and process quality from the fused multidimensional data matrix to generate a set of initial key features; S24: Perform dimensionality reduction and importance screening on the initial set of key features, eliminate redundant features, retain the feature subset that best represents the system state, and generate the final machine state feature vector.

[0009] Optionally, in step S22, an algorithm based on dynamic time warping is used to perform time series alignment on the preprocessed multivariate raw data, and a feature extractor based on a convolutional neural network is used to perform feature layer fusion, mapping the preprocessed multivariate raw data from different sources and types into a unified fused multidimensional data matrix.

[0010] Optionally, S3 includes: S31: Load the preset dynamic adjustment model from the storage unit of the edge computing node, and load the preset dynamic adjustment model into memory to complete the initialization; S32: Input the machine state feature vector into the preset dynamic adjustment model, complete the data format conversion and feature dimension matching, and generate a machine state feature vector that meets the model input requirements; S33: The preset dynamic adjustment model performs forward reasoning calculation on the machine state feature vector that meets the model input requirements, and generates preliminary grinding process parameter adjustment instructions through the built-in decision logic. S34: Perform process feasibility verification and output limiting processing on the preliminary grinding process parameter adjustment command to ensure that the command value is within the safe working range that the grinding robot can execute, and generate the final grinding process parameter adjustment command.

[0011] Optionally, the data format conversion includes converting the integer machine state feature vector into a floating-point tensor, and the feature dimension matching includes making the feature dimension consistent with the requirements of the model input layer by padding or truncation.

[0012] Optionally, S4 includes: S41: The final grinding process parameter adjustment instruction is encapsulated and format converted according to the industrial communication protocol specified by the bottom controller of the grinding robot to generate a grinding process parameter adjustment instruction data packet that conforms to the communication specifications of the bottom controller. S42: The grinding process parameter adjustment instruction data packet conforming to the underlying controller communication specification is sent to the grinding robot underlying controller in real time via industrial Ethernet or fieldbus. After the underlying controller parses and executes the instruction, it generates instruction execution status feedback containing instruction execution status and execution result. S43: Associate the instruction execution status feedback with the machine status feature vector and align the timestamps to generate a monitoring data packet containing complete execution process information; S44: The monitoring data packet containing complete execution process information is uploaded to the cloud platform through the IoT gateway. After receiving and parsing the data packet, the cloud platform performs real-time visualization and distributed storage to complete centralized monitoring and recording.

[0013] The IoT-based grinding robot monitoring system, used to implement the aforementioned IoT-based grinding robot monitoring method, includes the following modules: Sensor acquisition module: Used to collect diverse raw data of the polishing process in real time through an Internet of Things sensor network deployed on the polishing robot body and working environment; Edge processing module: Communicatively connected to the sensor acquisition module, used to transmit the multi-source raw data to the edge computing node for preprocessing and data fusion, and extract key features to generate machine state feature vector; Intelligent decision-making module: Located in the edge processing module, it is used to input the machine state feature vector into a preset dynamic adjustment model for analysis and calculation, and generate grinding process parameter adjustment instructions; Execution control module: Communicatively connected to the intelligent decision-making module, used to send the grinding process parameter adjustment instructions to the underlying controller of the grinding robot to perform real-time adjustments; Cloud service platform: Communicatively connected to the edge processing module and the execution control module, used to receive and store the machine status feature vector and grinding process parameter adjustment instructions, and realize centralized monitoring and recording.

[0014] An apparatus includes: at least one processor and a memory communicatively connected to said at least one processor; The memory stores instructions that can be executed by the at least one processor. When the instructions are executed by the at least one processor, the at least one processor is able to perform the IoT-based grinding robot monitoring method as described above.

[0015] The beneficial effects of this invention are: This invention, through the construction of an IoT sensor network, enables a grinding robot to simultaneously collect various raw signals, including vibration, acoustic emission, force / torque, and ambient temperature and humidity. Microsecond-level time alignment is achieved through a multi-channel synchronization mechanism (based on the IEEE 1588 protocol). By combining data cleaning, noise reduction, standardization, and CNN fusion modeling processes, not only is the representation quality of multidimensional state data effectively improved, but also a machine state feature vector that accurately reflects the robot's operational health and process quality is constructed through key feature extraction and dimensionality reduction filtering. This significantly enhances the system's real-time state perception and behavior modeling capabilities for complex grinding conditions.

[0016] This invention introduces a pre-defined dynamic adjustment model driven by machine state feature vectors, combined with a multi-layer neural network trained offline and a fuzzy expert system based on process rules, forming a hybrid decision logic that combines nonlinear expressiveness and interpretability. This model not only accurately outputs parameter adjustment suggestions for the current state, but also ensures the process rationality and physical feasibility of the output instructions through an expert rule correction mechanism. Furthermore, it applies a dead-zone limiting strategy during the output stage to effectively avoid instruction oscillations and control jitter, thereby improving the accuracy, stability, and on-site executability of parameter adjustments.

[0017] This invention, through an industrial protocol encapsulation, distribution, and feedback mechanism for adjusting final grinding process parameters, combined with local database data binding and a cloud-based data push process using TLS encryption, achieves a complete closed-loop control chain encompassing "status identification—decision generation—instruction execution—feedback verification—process archiving." The cloud platform, through WebSocket push and time-series database compression storage mechanisms, enables visualization, distributed recording, and long-term archiving management of the grinding process, providing highly reliable data support for subsequent process optimization, anomaly diagnosis, and operational decisions, thereby enhancing the system's intelligence level and the transparency of the production process. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the system flow according to an embodiment of the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0021] like Figure 1 As shown, the IoT-based monitoring method for polishing robots includes the following steps: S1: Through an IoT sensor network deployed on the polishing robot and its working environment, diverse raw data from the polishing process are collected in real time, specifically: S11: Initialize the IoT sensor network deployed on the grinding robot body and working environment, configure the sampling frequency and data format of each sensor, and generate an initialized and ready IoT sensor network.

[0022] First, vibration sensors, acoustic emission sensors, force / torque sensors, and ambient temperature and humidity sensors are installed on key structural parts of the grinding robot (such as the spindle box and arm) and their surrounding environmental control points. To achieve efficient data transmission, wired industrial Ethernet or wireless 5G communication links are constructed between each sensor and the edge computing node, and a reliable network address identification mechanism is configured. Then, based on the functional requirements and signal characteristics of each type of sensor, its physical address and logical identifier code are uniformly configured, and a device address mapping table is constructed for subsequent data identification and classification management. Subsequently, the sampling frequency of each type of sensor is uniformly set; for example, vibration and acoustic emission sensors are set to 10kHz, force / torque sensors to 1kHz, and ambient temperature and humidity sensors to 1Hz, and the data output format is set to 16-bit integer or 32-bit floating-point number format. Finally, the initialization verification of the IoT sensor network is completed, including communication link connectivity testing, synchronization signal reception testing based on the IEEE 1588 Precision Time Protocol (PTP), and online calibration of all sensors is performed using a standard signal source to confirm that their zero-point drift and sensitivity coefficient are within the allowable error range, marking the sensor network as initialized and ready.

[0023] S12: Through an initialized and ready IoT sensor network, synchronously collect initial sensor data streams of the grinding robot motor current, body vibration, grinding head acoustic emission, tool end force torque, and ambient temperature and humidity at a preset sampling frequency.

[0024] After initializing the IoT sensor network, a unified synchronization control signal triggering mechanism based on the IEEE 1588 Precision Time Protocol (PTP) is initiated to drive various sensors to perform high-precision synchronous acquisition according to a set sampling period (e.g., 1ms). Vibration sensors acquire the raw three-axis acceleration waveforms of the grinding robot's body structure during processing; acoustic emission sensors acquire high-frequency acoustic emission signals during the interaction between the grinding head and the workpiece; force / torque sensors acquire real-time force values ​​and instantaneous torque responses in three dimensions at the tool end; and environmental temperature and humidity sensors acquire ambient temperature and humidity readings of the working area. The acquisition results from these sensors form a structured initial sensor data stream according to the sampling time window and are written to the input buffer of the edge computing node in real time through their respective data channels. This buffer uses a ping-pong caching mechanism to decouple data acquisition from subsequent processing, ensuring continuous, stable, and lossless acquisition.

[0025] S13: Perform real-time verification and timestamp synchronization on the initial sensor data stream to eliminate data packet loss and timing disorder issues, and generate synchronized and uniformly formatted multi-source raw data.

[0026] Upon receiving the initial sensor data stream, the edge computing node first performs an integrity check on each data frame uploaded by the sensor. It uses the CRC-32 algorithm to calculate the checksum of the data packets and compares it with the header and footer identifier fields. If a data packet is found to be missing, fragmented, or has an error in an abnormal field, the corresponding data segment is immediately removed to ensure the accuracy of subsequent processing.

[0027] Based on the IEEE 1588 protocol, which shares the same origin as the acquisition synchronization signal, a uniform microsecond-level timestamp is appended to all valid data to ensure strict alignment between multi-source data. After time alignment, for low-frequency signals such as ambient temperature and humidity, a linear interpolation algorithm is used to map them to a unified timeline reference for high-frequency data, and all data fields are standardized to a uniform 32-bit floating-point format. The final output data is a synchronized and uniformly formatted set of multi-source raw data, providing a foundation for subsequent framing and analysis.

[0028] S14: The synchronized and uniformly formatted multi-source raw data is framed and buffered according to the preset data packet size to generate multi-source raw data to be transmitted.

[0029] After synchronization and format unification, edge computing nodes perform structured framing of various types of raw data according to the MQTT application layer protocol's encapsulation standards and within a fixed time window (e.g., 500ms). During encapsulation, metadata tags are added to each sensor data segment, including sampling timestamps, sensor logical addresses, channel numbers, and data quality identifiers (e.g., CRC check results). This structured encapsulation ensures that each data packet has complete semantics, accurate origin, and stable identifiability. The framed data packets are stored in the edge node's high-performance circular buffer. The buffer implements real-time monitoring of data write status and overflow protection mechanisms to ensure data integrity and controllability before transmission. The final result is raw data with a unified structure, complete timing, and conformity to transmission protocol standards, ready for transmission.

[0030] S2: The raw multi-source data is transmitted to edge computing nodes for preprocessing and data fusion, and key features are extracted to generate a machine state feature vector representing the current overall state, specifically: S21: The edge computing node receives the multi-dimensional raw data to be transmitted, and performs data cleaning, noise reduction and standardization on the multi-dimensional raw data to be transmitted to generate pre-processed multi-dimensional raw data.

[0031] The edge computing node first reads the multi-dimensional raw data to be transmitted, which is encapsulated in time windows from the cache. This data includes the original vibration waveform, the original acoustic emission waveform, the original triaxial force / torque data, and the ambient temperature and humidity values.

[0032] For vibration and acoustic emission data, wavelet thresholding denoising is performed: the signal is decomposed into multi-scale wavelets, the Daubechies-4 wavelet basis is selected, the adaptive soft thresholds for each scale are calculated, and the wavelet coefficients are compressed and reconstructed, thereby effectively suppressing high-frequency noise and preserving the main structure of the signal.

[0033] For force / torque data, a moving average filtering method is applied: a weighted average is processed with a fixed window length (e.g., 10 points) to smooth short-term jitter and improve the observability of fluctuation trends.

[0034] Subsequently, Z-Score standardization was performed on all categories of data: the mean and standard deviation of each group of data were calculated, and then the mean was subtracted from each data point and divided by the standard deviation to unify the numerical distribution, ensuring that the data in each channel have a uniform dimension and scale, and finally generating preprocessed multivariate raw data.

[0035] S22: Perform spatiotemporal alignment and feature layer fusion on the preprocessed multivariate raw data to map the preprocessed multivariate raw data from different sources and types to a unified high-dimensional feature space, generating a fused multidimensional data matrix.

[0036] To achieve a unified temporal and semantic representation of multi-channel data, the following two processes are executed first: Time alignment: Due to differences in sampling start points and response delays in the preprocessed data of each channel, an alignment algorithm based on dynamic time warping is adopted. Taking the main vibration channel as a reference, the shortest path matching is calculated for each channel, and the non-reference channel sequences are dynamically compressed or stretched to ensure that the multi-source data are aligned on a unified time axis.

[0037] Feature fusion: Preprocessed, time-aligned multivariate raw data is input into a feature extractor constructed from a one-dimensional convolutional neural network. This network contains multi-scale one-dimensional convolutional kernels to extract local short-term features and global long-range correlation features, respectively. The extracted features are integrated through a feature concatenation layer and then subjected to batch normalization to finally form a fused multidimensional data matrix with consistent structure and uniform dimensions. Each row of this matrix corresponds to the sensor feature representation at a unified time step, and each column corresponds to the projection dimension of a sensor type in high-dimensional space.

[0038] S23: Extract time-domain, frequency-domain, and time-frequency-domain indicators that are strongly correlated with equipment status and process quality from the fused multidimensional data matrix to generate a set of initial key features.

[0039] For different sensing channels in the fused multidimensional data matrix, key indicators are extracted using the following method. All features are calculated based on a sliding time window, with the center timestamp of the sliding window used as the feature label time point.

[0040] Vibration channel: Extract its effective value and kurtosis in the time domain, and calculate the segmented spectral energy values ​​of its first, second and third harmonics in the frequency domain by Fast Fourier Transform (FFT).

[0041] Acoustic emission channel: Extract its ring count and root mean square value to capture local impact events such as microcracks and their energy levels during the grinding process.

[0042] Force / Torque Channel: Extracts the maximum value and fluctuation variance of the force in three dimensions to quantify the applied force intensity and its stability.

[0043] The key feature indicators from each channel are integrated and combined to form a set of initial key features, which are used for subsequent dimensionality reduction and screening.

[0044] S24: Perform dimensionality reduction and importance screening on a set of initial key features, eliminate redundant features, retain the feature subset that best represents the system state, and generate the final machine state feature vector.

[0045] This step is divided into two stages: Dimensionality reduction: Principal component analysis algorithm is used to linearly reduce the dimensionality of the above set of initial key features. By constructing the feature covariance matrix and performing eigenvalue decomposition, the principal component dimensions with a cumulative variance contribution rate of more than 90% are extracted, ensuring that the data structure is simplified while retaining key information.

[0046] Feature selection: The dimensionality-reduced feature set is input into a random forest model trained offline using historical multi-condition refined samples. The Gini importance score of each feature is calculated and sorted according to the score. Based on the comprehensive requirements of system computing power and state representation accuracy, the top N features (e.g., N=12) are selected to form a feature subset with high information density, low redundancy, and clear physical meaning.

[0047] The final generated feature subset is the final machine state feature vector, which will serve as the input basis for dynamic adjustment of subsequent grinding process parameters.

[0048] S3: Input the machine state feature vector into the preset dynamic adjustment model for analysis and calculation, and generate grinding process parameter adjustment instructions for real-time optimization of the grinding process, specifically: S31: Load the preset dynamic adjustment model from the storage unit of the edge computing node and load the preset dynamic adjustment model into memory to complete the initialization.

[0049] After receiving the final machine state feature vector, the edge computing node first retrieves a pre-defined dynamically adjusted model from its local storage. This model is a machine learning model trained offline by a cloud-based training platform using historical polishing data and then distributed to the edge node. During loading, the edge computing node first performs model integrity and version consistency checks. Specifically, this includes calculating the hash digest of the model's compressed package using the SHA-256 algorithm and comparing it one-to-one with a pre-stored signature file to verify the authenticity and integrity of the model file. Then, it calls a built-in symmetric encryption module (e.g., using AES-256 encryption) to decrypt the model parameter file, restoring the structure diagram and parameter weight data of the pre-defined dynamically adjusted model. Once loading is complete, the pre-defined dynamically adjusted model is loaded into the edge node's memory, and the runtime environment initialization and inference engine registration are completed, placing it in a ready state to receive input data for inference computation.

[0050] S32: Input the machine state feature vector into the preset dynamic adjustment model, complete the data format conversion and feature dimension matching, and generate a machine state feature vector that meets the model input requirements.

[0051] To ensure the machine state feature vector can be correctly parsed by the pre-defined dynamically adjusted model, edge computing nodes need to perform standardized preprocessing. First, a data format conversion operation is performed, converting the original set of integer feature values ​​item by item into a 32-bit floating-point data array to meet the input precision requirements of the deep learning model. Then, based on the structural definition of the model's input layer parameters, the converted data array is organized into a standard multi-dimensional array structure to construct a data carrier that matches the model's input interface. On this basis, a feature dimension matching operation is further performed: if the current feature vector dimension is less than the model's required input dimension, zero-value elements are added sequentially to pad the tail; if the feature vector dimension exceeds the model's input requirements, the first few dimensions are truncated according to a pre-defined importance ranking rule, and redundant elements at the end are discarded. The final output is a machine state feature vector whose data type and dimensional structure both meet the model's requirements, and can be directly used for subsequent model inference calculations.

[0052] S33: The preset dynamic adjustment model performs forward reasoning calculations on the machine state feature vectors that meet the model input requirements, and generates preliminary grinding process parameter adjustment instructions through the built-in decision logic.

[0053] After the machine state feature vector that meets the model input requirements is input, the edge computing node immediately invokes the preset dynamic adjustment model for forward inference calculation. This model adopts a hybrid decision-making mechanism that integrates a neural network model with a fuzzy rule-based expert system. While balancing nonlinear expressiveness and interpretable control constraints, it achieves more reliable process parameter adjustment. The neural network model is a multi-layer feedforward neural network structure containing multiple hidden layers and nonlinear activation functions, which can effectively learn the complex nonlinear mapping relationship between high-dimensional input and continuous output, and generate a set of preliminary process parameter recommendations. The expert system is based on a fuzzy rule reasoning framework. It receives parameters from the neural network as input and performs boundary correction and safety limits by searching for fuzzy control rules in the process knowledge base (such as "if the workpiece vibration value is high and the feed speed is too large, then reduce the spindle speed").

[0054] This hybrid decision-making mechanism effectively compensates for the "black box" nature of neural network models and the limitations of overly rigid static rules in expert systems, combining adaptability with control rigidity. Finally, by synthesizing the above calculation results, a set of preliminary grinding process parameter adjustment instructions, including multiple variables such as spindle speed, feed rate, and grinding pressure, is generated.

[0055] S34: Perform process feasibility verification and output limiting processing on the preliminary grinding process parameter adjustment instructions to ensure that the instruction values ​​are within the safe working range that the grinding robot can execute, and generate the final grinding process parameter adjustment instructions.

[0056] To ensure that the initial grinding process parameter adjustment command does not lead to control anomalies or equipment operation risks during actual issuance, the edge computing node performs a series of security verification operations on the command. First, it calls the grinding process parameter database to obtain the minimum and maximum allowed values ​​of the current process parameters, and compares them item by item with the command value to complete the process feasibility verification.

[0057] For cases where the instruction value exceeds the boundary but is still within the error tolerance range, instead of directly performing hard truncation, a limiting algorithm with a dead zone is used for flexible limitation: a dynamic limiting threshold band is set below the upper limit of the parameter (for example, setting the dead zone boundary at 3% of the upper limit). When the instruction value enters the dead zone, no immediate limiting operation is performed. It is only fixed to the safe upper limit when it exceeds the threshold, thus avoiding frequent instruction jumps caused by external disturbances or feedback oscillations.

[0058] Compared to the traditional boundary truncation method, it is more robust and can effectively prevent frequent fluctuations in the execution system under extreme conditions, thus improving control stability. All control commands after amplitude limiting correction constitute the final grinding process parameter adjustment commands, which will be issued in real time as the execution targets of the subsequent S4 step.

[0059] S4: The grinding process parameter adjustment command is sent to the underlying controller of the grinding robot for real-time adjustment. At the same time, the machine status feature vector and the grinding process parameter adjustment command are uploaded to the cloud platform for centralized monitoring and recording. Specifically: S41: The final grinding process parameter adjustment command is encapsulated and format converted according to the industrial communication protocol specified by the underlying controller of the grinding robot, generating a grinding process parameter adjustment command data packet that conforms to the communication specifications of the underlying controller.

[0060] The edge computing node first standardizes the format of the final grinding process parameter adjustment instructions according to the industrial communication protocols (such as EtherCAT or PROFINET real-time industrial Ethernet protocols) supported by the underlying controller of the grinding robot. During the encapsulation process, the system performs byte-aligned encoding on each parameter item (such as spindle speed, feed rate, grinding pressure, etc.) according to the data structure requirements specified by the protocol to ensure the compliance of the data frame and the parsing efficiency.

[0061] Subsequently, message header information required by industrial communication protocol standards is added, including fields such as target device address, function code, channel identifier, and data length. A verification module (such as CRC-32) is invoked to generate a checksum field for the complete data segment, ensuring data integrity and anti-interference capabilities during transmission. A timestamp field under the current system's unified clock is also appended to facilitate time synchronization and traceability analysis of subsequent execution status. These fields are encapsulated and combined according to the protocol format to form a grinding process parameter adjustment instruction data packet conforming to the underlying controller's communication specifications.

[0062] S42: The grinding process parameter adjustment instruction data packet conforming to the communication specification of the underlying controller is sent to the underlying controller of the grinding robot in real time via industrial Ethernet or fieldbus. After the underlying controller parses and executes the instruction, it generates instruction execution status feedback containing instruction execution status and execution result.

[0063] Edge computing nodes transmit data packets to the grinding robot's underlying controller in real time via industrial Ethernet or fieldbus. Upon receiving the instruction data packet, the underlying controller first performs an integrity verification process to check whether the message header, data length, and checksum field match.

[0064] If the verification passes, the controller immediately extracts the instruction parameter values ​​from the data body and writes each parameter into the corresponding control register address in the robot motion control loop, such as the speed loop register, pressure regulation module address, and other control channels, according to the pre-configured address mapping relationship, so as to drive the robot's underlying control module to complete the corresponding action.

[0065] After the task is completed, the underlying controller encapsulates the instruction execution status (success, failure, error type code) and the actual execution results of each process parameter into instruction execution status feedback, and returns it to the edge computing node through the same bus, forming the basic control feedback mechanism of the system and providing real-time basis for closed-loop adaptive optimization.

[0066] S43: Associate the instruction execution status feedback with the machine status feature vector and align the timestamps to generate a monitoring data packet containing complete execution process information.

[0067] After receiving instruction execution status feedback, the edge computing node needs to structurally associate it with the machine state feature vector on which the instruction was generated, in order to achieve full-process data traceability, comparability, and analyzability. To achieve data association and storage, the node internally calls a built-in lightweight relational database management system to establish a key-value association between the executed instruction and the corresponding machine state feature vector, using the instruction number as the primary key, ensuring that every control action can be traced back to the state information at that time.

[0068] Furthermore, based on the IEEE 1588 precision time protocol, the system performs microsecond-level alignment between the feedback data and the timestamps attached to the feature vectors, uniformly mapping them onto a standard time axis to ensure the synchronization of data across channels. After integrating command inputs, feature states, feedback execution results, and time information, a complete and auditable monitoring data package containing comprehensive execution process information is generated for subsequent uploading and analysis.

[0069] S44: The monitoring data packet containing complete execution process information is uploaded to the cloud platform through the IoT gateway. After receiving and parsing the data packet, the cloud platform performs real-time visualization and distributed storage to complete centralized monitoring and recording.

[0070] Once the monitoring data packet is constructed at the edge node, it is immediately uploaded to the cloud platform through the deployed IoT gateway module. The data upload channel is established based on a lightweight messaging protocol (such as MQTT) using a publish / subscribe model. During communication, a TLS (Transport Layer Security) encrypted channel is used to achieve encrypted data transmission and authentication, ensuring communication security.

[0071] After receiving the monitoring data packet, the cloud platform first performs protocol desealing, field parsing, and data verification to extract core fields such as feature vectors, command parameters, execution feedback, and timestamps.

[0072] In terms of real-time visualization, the platform's front end establishes a two-way communication channel with the back end through the WebSocket protocol. It uses a push mechanism to dynamically render core monitoring data in the form of charts on the web page, including parameter curves, execution feedback status, operation logs, etc., and supports multi-channel overlay display and abnormal highlighting.

[0073] In terms of distributed storage, the platform backend calls a database engine dedicated to time-series data storage, such as a time-series database that uses a columnar compression structure and an efficient indexing mechanism to partition and compress various monitoring data according to dimensions such as time, robot ID, and parameter type, supporting high-concurrency writing and multi-dimensional historical data retrieval.

[0074] Ultimately, a seamless connection from edge computing to the cloud platform was achieved, completing the closed-loop control link of grinding process parameters from "perception-reasoning-execution-feedback-monitoring". A highly reliable, highly visible and highly traceable intelligent manufacturing monitoring system was built, providing a solid data foundation for subsequent anomaly identification, process optimization and big data analysis.

[0075] like Figure 2 As shown, the IoT-based grinding robot monitoring system is used to implement an IoT-based grinding robot monitoring method, and includes the following modules: Sensor acquisition module: Used to collect diverse raw data of the polishing process in real time through an Internet of Things sensor network deployed on the polishing robot body and working environment; Edge processing module: It communicates with the sensor acquisition module and is used to transmit multi-source raw data to the edge computing node for preprocessing and data fusion, extract key features, and generate machine state feature vectors; Intelligent decision-making module: Located in the edge processing module, it is used to input the machine state feature vector into the preset dynamic adjustment model for analysis and calculation, and generate grinding process parameter adjustment instructions; Execution control module: Communicates with the intelligent decision-making module and is used to send grinding process parameter adjustment instructions to the underlying controller of the grinding robot to perform real-time adjustments; Cloud service platform: Communicates with edge processing module and execution control module to receive and store machine status feature vectors and grinding process parameter adjustment instructions, enabling centralized monitoring and recording.

[0076] The present invention also includes related devices, including: at least one processor and a memory communicatively connected to at least one processor; The memory stores instructions that can be executed by at least one processor. When executed by at least one processor, the instructions enable at least one processor to perform the IoT-based grinding robot monitoring method described above.

[0077] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0078] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring grinding robots based on the Internet of Things, characterized in that, Includes the following steps: S1: Collect diverse raw data of the polishing process in real time through an IoT sensor network deployed on the polishing robot body and working environment; S2: The multi-source raw data is transmitted to the edge computing node for preprocessing and data fusion, and key features are extracted to generate a machine state feature vector that represents the current comprehensive state. S3: Input the machine state feature vector into a preset dynamic adjustment model for analysis and calculation, and generate grinding process parameter adjustment instructions for real-time optimization of the grinding process; S4: Send the grinding process parameter adjustment command to the bottom controller of the grinding robot to perform real-time adjustment, and at the same time upload the machine status feature vector and the grinding process parameter adjustment command to the cloud platform for centralized monitoring and recording.

2. The IoT-based grinding robot monitoring method according to claim 1, characterized in that, S1 includes: S11: Initialize the IoT sensor network deployed on the grinding robot body and working environment, configure the sampling frequency and data format of each sensor, and generate an initialized and ready IoT sensor network. S12: Through the initialized IoT sensor network, the initial sensor data streams of the grinding robot motor current, body vibration, grinding head acoustic emission, tool end force torque and ambient temperature and humidity are synchronously collected at a preset sampling frequency. S13: Perform real-time verification and timestamp synchronization on the initial sensing data stream to eliminate data packet loss and timing disorder, and generate synchronized and uniformly formatted multi-source raw data; S14: The synchronized and uniformly formatted multi-source raw data is framed and buffered according to a preset data packet size to generate multi-source raw data to be transmitted.

3. The IoT-based grinding robot monitoring method according to claim 2, characterized in that, The IoT sensor network includes the physical addresses and logical mapping relationships of vibration sensors, acoustic emission sensors, force / torque sensors, and environmental temperature and humidity sensors, and uniformly sets the sampling frequency and data output format of each sensor to complete the generation of the initialized and ready IoT sensor network.

4. The IoT-based grinding robot monitoring method according to claim 3, characterized in that, S2 includes: S21: The edge computing node receives the multi-source raw data to be transmitted, and performs data cleaning, noise reduction and standardization processing on the multi-source raw data to be transmitted to generate preprocessed multi-source raw data. S22: Perform spatiotemporal alignment and feature layer fusion on the preprocessed multivariate raw data to map the preprocessed multivariate raw data from different sources and types to a unified high-dimensional feature space, and generate a fused multidimensional data matrix. S23: Extract time-domain, frequency-domain, and time-frequency-domain indicators that are strongly correlated with equipment status and process quality from the fused multidimensional data matrix to generate a set of initial key features; S24: Perform dimensionality reduction and importance screening on the initial set of key features, eliminate redundant features, retain the feature subset that best represents the system state, and generate the final machine state feature vector.

5. The IoT-based grinding robot monitoring method according to claim 4, characterized in that, In step S22, an algorithm based on dynamic time warping is used to align the preprocessed multivariate raw data to a time series, and a feature extractor based on a convolutional neural network is used to perform feature layer fusion, mapping the preprocessed multivariate raw data from different sources and types into a unified fused multidimensional data matrix.

6. The IoT-based monitoring method for grinding robots according to claim 5, characterized in that, S3 includes: S31: Load the preset dynamic adjustment model from the storage unit of the edge computing node, and load the preset dynamic adjustment model into memory to complete the initialization; S32: Input the machine state feature vector into the preset dynamic adjustment model, complete the data format conversion and feature dimension matching, and generate a machine state feature vector that meets the model input requirements; S33: The preset dynamic adjustment model performs forward reasoning calculation on the machine state feature vector that meets the model input requirements, and generates preliminary grinding process parameter adjustment instructions through the built-in decision logic. S34: Perform process feasibility verification and output limiting processing on the preliminary grinding process parameter adjustment command to ensure that the command value is within the safe working range that the grinding robot can execute, and generate the final grinding process parameter adjustment command.

7. The IoT-based grinding robot monitoring method according to claim 6, characterized in that, The data format conversion includes converting integer machine state feature vectors into floating-point tensors, and feature dimension matching includes padding or truncation to make the feature dimensions consistent with the requirements of the model input layer.

8. The IoT-based grinding robot monitoring method according to claim 7, characterized in that, S4 includes: S41: The final grinding process parameter adjustment instruction is encapsulated and format converted according to the industrial communication protocol specified by the bottom controller of the grinding robot to generate a grinding process parameter adjustment instruction data packet that conforms to the communication specifications of the bottom controller. S42: The grinding process parameter adjustment instruction data packet conforming to the underlying controller communication specification is sent to the grinding robot underlying controller in real time via industrial Ethernet or fieldbus. After the underlying controller parses and executes the instruction, it generates instruction execution status feedback containing instruction execution status and execution result. S43: Associate the instruction execution status feedback with the machine status feature vector and align the timestamps to generate a monitoring data packet containing complete execution process information; S44: The monitoring data packet containing complete execution process information is uploaded to the cloud platform through the IoT gateway. After receiving and parsing the data packet, the cloud platform performs real-time visualization and distributed storage to complete centralized monitoring and recording.

9. An IoT-based grinding robot monitoring system, used to implement the IoT-based grinding robot monitoring method as described in any one of claims 1-8, characterized in that, Includes the following modules: Sensor acquisition module: Used to collect diverse raw data of the polishing process in real time through an Internet of Things sensor network deployed on the polishing robot body and working environment; Edge processing module: Communicatively connected to the sensor acquisition module, used to transmit the multi-source raw data to the edge computing node for preprocessing and data fusion, and extract key features to generate machine state feature vector; Intelligent decision-making module: Located in the edge processing module, it is used to input the machine state feature vector into a preset dynamic adjustment model for analysis and calculation, and generate grinding process parameter adjustment instructions; Execution control module: Communicatively connected to the intelligent decision-making module, used to send the grinding process parameter adjustment instructions to the underlying controller of the grinding robot to perform real-time adjustments; Cloud service platform: Communicatively connected to the edge processing module and the execution control module, used to receive and store the machine status feature vector and grinding process parameter adjustment instructions, and realize centralized monitoring and recording.

10. A device, characterized in that, include: At least one processor and a memory communicatively connected to said at least one processor; The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform the IoT-based grinding robot monitoring method as described in any one of claims 1 to 8.

Images