Intelligent decoration wall flatness laser detection and automatic leveling system

Abstract

The application discloses a kind of intelligent decoration wall surface flatness laser detection and automatic leveling system, it is related to building decoration intelligent construction technical field, the system includes: acquisition control module is marked data according to process acquisition, multi-source fusion generates standard data set upload cloud encryption warehousing, parameter calculus module federal learning training stratification and cross-process model, cross-stage correction reference parameter is issued edge, edge setting module closed-loop fine-tuning generates execution parameter, data double-path retention feedback, link exception time offline model bottom, after recovery, middle station feedback iteration optimization acquisition and model configuration;The application adopts stratified data acquisition and multi-source feature fusion, terminal data is not transmitted, only model parameter is interacted, cloud federal learning continues to optimize leveling reference, edge end fine-tuning in situ cooperates offline bottom, data local and cloud double-path retention closed-loop reuse, network exception time lightweight model offline operation, after recovery, fill transmission iteration, improve intelligent leveling precision and on-site reliability.

Description

Technical Field

[0001] This invention relates to the field of intelligent construction technology for building decoration, specifically to an intelligent laser detection and automatic leveling system for the flatness of decorative walls. Background Technology

[0002] With the accelerating industrialization of the building decoration industry, high-end interior decoration projects are placing higher demands on the quality and efficiency of wall leveling. Wall leveling typically involves multiple processes, including rough leveling, fine leveling, and precision leveling, and its quality directly affects the subsequent application of finishing layers and paint. In recent years, technologies such as laser ranging, industrial internet, edge computing, and artificial intelligence have gradually penetrated the building construction field, driving the transformation of wall leveling from reliance on manual experience to digitalization and intelligence. How to integrate multi-source construction data, achieve collaborative optimization of parameters between processes, and ensure the continuity and reliability of on-site construction has become a key area of ​​technological research and development in this field.

[0003] The paper "A wall plastering machine and its leveling method" with publication number CN104631804A achieves mechanical angle positioning of the plastering plate through a light and shadow projection device and a receiving device. However, it can only complete simple photoelectric sensing leveling at the level of a single machine. It lacks the ability to collect and integrate multi-source construction data, and does not involve data cloud storage, parameter model calculation or edge closed-loop control. It cannot adapt to wall leveling scenarios with multiple processes and multiple factors coupled. The patent application CN120374308A, titled "High-Precision Construction Method and System for Large-Area Flooring Based on BIM Model," introduces a federated learning framework to optimize compartmentalization strategies and uses blockchain for data storage and digital twin management. However, its technical solution is specifically designed for large-area flooring concrete pouring construction and does not involve a multi-stage, layered prediction model for wall leveling. It also does not provide cross-stage error compensation, edge-end closed-loop fine-tuning, or offline fallback mechanisms for network anomalies. The application objects of its federated learning are fundamentally different from the process requirements of wall leveling and cannot be directly transferred to the intelligent control of rough, fine, and precise wall leveling.

[0004] In current wall leveling construction, most solutions still rely on the experience of construction workers to determine process parameters such as mortar spreading thickness and trowel speed. Laser inspection equipment is only used as a single tool for measuring flatness and cannot simultaneously collect related information such as environment, raw materials, and defects in previous processes, resulting in the inability to integrate and analyze multi-source data. The parameters of each leveling process are set independently, lacking the ability to transmit and compensate for construction errors in previous processes, which easily leads to the accumulation of errors at each stage. In addition, construction data is mostly recorded in paper ledgers, making it difficult to form a large-scale process sample library, and process optimization relies on manual trial and error; equipment control is highly dependent on continuous cloud network connectivity. Once the communication link is abnormal, the intelligent equipment cannot perform autonomous calculations and must revert to manual operation. Moreover, the parameters sent from the cloud cannot be dynamically adjusted according to real-time deviations on site, lacking closed-loop control logic, ultimately resulting in large fluctuations in wall flatness and difficulty in ensuring construction reliability. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent laser detection and automatic leveling system for the flatness of interior walls. The system aggregates multi-dimensional construction information through hierarchical acquisition, forms a standardized dataset through multi-source feature fusion, and performs data encryption, cleaning, and classification storage in the cloud. Relying on federated learning, the system prevents the transmission of raw local data, builds process prediction models step-by-step, optimizes construction baseline parameters using cross-order correction algorithms, and fine-tunes parameters at the edge based on on-site deviations, storing construction data in partitions. Data is retained locally and transmitted back to the cloud, enabling data reuse. A link maintenance module monitors network status in real time, and in case of communication failure, a lightweight local model is activated for offline operation. Combined with a middleware service, the system continuously iterates the acquisition rules and model configurations. The entire solution achieves digital management of the entire wall leveling process, improves leveling accuracy and equipment stability, adapts to various decoration construction scenarios, and facilitates the intelligent transformation and upgrading of the wall leveling process.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an intelligent laser detection and automatic leveling system for the flatness of decorative walls, including a data acquisition and control module, a cloud database module, a parameter calculation module, an edge calibration module, and a link maintenance module.

[0007] The data acquisition and control module is used to collect construction data of coarse leveling, fine leveling and precision leveling processes according to specifications and mark the working conditions. It uses a multi-source feature fusion algorithm to generate a standardized construction feature dataset and sends it to the cloud storage module. The cloud-based data entry module is used to receive and encrypt the dataset through the digital industrial internet, clean it, and then store it in the cloud-based process knowledge base and federated learning sample pool according to the process. The parameter calculation module is used to retrieve data from the knowledge base and sample pool, use federated learning to train a hierarchical prediction model with a convolutional neural network architecture to predict single-process benchmark parameters, train a cross-process linkage prediction model to calculate the impact compensation value of the preceding process on the subsequent process, use a cross-order benchmark correction algorithm to correct the benchmark parameters using the impact compensation value, and generate an instruction package to be sent to the edge tuning module via the industrial internet. The edge tuning module is used to receive the instruction packet, and uses a field closed-loop fine-tuning algorithm to fine-tune the reference parameters based on the real-time deviation. The execution parameters are generated by PID control and sent to the automatic wall leveling execution mechanism. The automatic wall leveling execution mechanism completes the wall leveling construction operation of the corresponding process according to the execution parameters. During the construction process, the flatness, working conditions and equipment operation data collected in real time are uniformly cached to the local storage unit at the edge to form cache data. After the construction, the cache data is stored and sent back in two ways to switch the control logic of the next process. The link operation and maintenance module consists of a latency detection unit, a local model switching unit, an offline data collection unit, and a middleware interaction unit. It is used to monitor communication link latency. When latency is abnormal, it calls the local lightweight flatness prediction model and uses cached data for offline parameter calculation instead of cloud calculation. After recovery, it uploads offline data and relies on the middleware to provide diagnostic, recommendation, and update services. It also adjusts the collection rules and model configuration based on feedback.

[0008] Furthermore, when the data acquisition and control module performs the data acquisition operations for the coarse leveling, fine leveling, and precision leveling processes, the coarse leveling completes data acquisition at a fixed low-frequency cycle of once every 5-10 seconds, the fine leveling completes data acquisition at a continuous medium-frequency cycle of once every 1-2 seconds, and the precision leveling completes data acquisition at a dense high-frequency cycle of more than 10 times per second. The construction data entered during the acquisition process includes wall flatness data, base surface type identification data, environmental temperature and humidity data, mortar performance data, and defect marking data of previous processes. All entered data is bound to the corresponding working condition identification content.

[0009] Furthermore, the process of the cloud-based data entry module receiving standardized construction feature datasets is divided into two transmission periods: during the construction operation period, the dataset is split into multiple sub-data packets and transmitted in segments with encryption; during the downtime period, the entire dataset is collected, packaged, and transmitted with encryption. The data cleaning operation of the cloud-based data entry module removes blank fields and abnormal content with values ​​exceeding the reasonable range of working conditions. The remaining content is stored in the corresponding partition of the cloud-based process knowledge base and the federated learning sample pool according to the division rules of coarse leveling, fine leveling, and precise leveling.

[0010] Furthermore, the federated learning training process of the parameter calculation module follows a fixed data interaction specification. The original construction data collected by all data acquisition terminals is stored locally on the terminal and does not participate in network transmission. The data acquisition terminal only transmits model gradient iteration data and process statistics data to the parameter calculation module. The model gradient iteration data includes the gradient update values ​​of process parameters and feature weights for each process. The process statistics data includes single-process construction condition label data and process feature aggregation data. After the parameter calculation module summarizes and aggregates the model gradient iteration data uploaded by all terminals, it matches the coarse leveling process, fine leveling process, and precision leveling process respectively, and independently completes the iterative training of the dedicated hierarchical prediction model for each process. The hierarchical prediction model for each process adopts a network architecture of input layer, two-level feature extraction hidden layer, and output layer. The dimension of the input layer is consistent with the feature dimension of the standardized construction feature dataset. The two-level hidden layers are set with 64 and 32 convolutional kernels respectively to extract multi-source construction features. The output layer outputs the benchmark parameters of the corresponding process, including three core control parameters: mortar paving thickness, scraper travel speed, and scraper pressing amount.

[0011] Furthermore, the parameter calculation module matches independent flatness value limits for each of the coarse leveling layer prediction model, fine leveling layer prediction model, and precision leveling layer prediction model. The coarse leveling layer prediction model is bound to a flatness value limit range of -2mm to +2mm, the fine leveling layer prediction model is bound to a flatness value limit range of -1mm to +1mm, and the precision leveling layer prediction model is bound to a flatness value limit range of -0.5mm to +0.5mm. After the cross-order benchmark correction algorithm completes a single parameter calculation, it immediately matches the corrected calculation parameters to the flatness of the current construction process. The numerical comparison is completed within the specified range of integer flatness values. If the corrected calculation parameter value is within the specified range of flatness values ​​for the corresponding process, the current corrected calculation parameter is retained. If the corrected calculation parameter value exceeds the specified range of flatness values ​​for the corresponding process, the current corrected calculation parameter is discarded, and the standardized construction feature dataset, benchmark parameters, and impact compensation values ​​are retrieved again. The parameter calculation process of the cross-level benchmark correction algorithm is then executed again. If the calculation results still exceed the specified range after three consecutive calculations, the valid calculation parameter that passed the previous verification in the current process is used as the current calculation parameter, and an alarm message is generated.

[0012] Furthermore, the edge adjustment module records all construction data within the complete construction cycle of a single leveling process. The complete construction cycle covers all time-series data from the start time to the end time of the process. The edge adjustment module retrieves all construction data collected within the complete construction cycle and matches the flatness limit value range corresponding to the current construction process to complete the data comparison group by group. After all time-series construction data matches the flatness limit value range of the corresponding process, the edge adjustment module performs a cache storage operation on the full amount of construction data.

[0013] Furthermore, the construction data cached by the edge tuning module is processed independently in two paths. The first data link continuously stores the cached construction data in a dedicated local storage partition at the edge. When the next process starts data acquisition, the cached data in the dedicated local storage partition at the edge serves as historical working condition feature parameters and participates in the construction of the standardized acquisition dataset for the next process. The second data link packages the cached construction data according to the unified encapsulation format of the digital industrial internet and transmits it to the cloud storage module via the uplink encrypted channel of the digital industrial internet. After the two data links are completed, the edge tuning module calls the preset process logic configuration file to complete the switching operation from the current process to the next process control logic.

[0014] Furthermore, the link operation and maintenance module is configured with a fixed communication latency judgment threshold of 100ms. It continuously acquires the real-time latency values ​​of the uplink and downlink of the digital industrial internet according to a fixed time sequence acquisition frequency, and completes the comparison operation between the real-time latency value and the preset communication latency judgment threshold. When the real-time latency value is greater than 100ms or there is a continuous data interruption in the uplink and downlink, it generates a local model call instruction and reads the lightweight flatness prediction model program file pre-stored locally at the edge.

[0015] Furthermore, after the link maintenance module identifies that the digital industrial internet communication link has switched from an abnormal state to a normal connectivity state, it locks the complete timestamp interval corresponding to the abnormal working condition of the link, collects the construction data, offline status parameter calculation records, and offline working condition parameter fine-tuning coefficient data within the timestamp interval, performs unified format normalization processing on all collected data, and uploads the normalized complete data file in batches through the digital industrial internet uplink channel, with the upload target location being the cloud storage module.

[0016] Furthermore, the link operation and maintenance module invokes three independent digital service units provided by the digital industrial internet platform: diagnostic service, recommendation service, and update service. The diagnostic service unit performs item-by-item verification and recording operations on the entire link time-series transmission data, model calculation data, and process parameter data. The recommendation service unit matches the corresponding parameter dataset within the cloud-based process knowledge base based on real-time operating condition tags. The update service unit generates configuration modification forms for model weight parameters and process constraint parameters. It summarizes all ledger data generated by the operation of the three types of digital service units. Based on the ledger data, the link operation and maintenance module revises the data acquisition timing rules and data annotation rules of the acquisition and control module, as well as the model iteration training parameters and threshold configuration parameters of the parameter calculation module.

[0017] Compared with existing technologies, this intelligent wall surface flatness laser detection and automatic leveling system has the following advantages: I. This invention utilizes a hierarchical data acquisition architecture coupled with a multi-source feature fusion processing mode. It integrates multi-dimensional construction influencing factors based on differentiated acquisition logic for different leveling processes. Combined with cloud-terminal separation data transmission rules and federated learning operation logic, the original construction data is stored locally on the terminal and not transmitted externally. Only related data is iterated through interactive models. This comprehensively summarizes information related to the environment, materials, and process defects in wall construction across all dimensions, forming a unified and standardized dataset. It also avoids the risk of leakage of original construction information from the data source. Data is organized, classified, and stored using a cloud-based knowledge base and sample pool. A hierarchical model adapts to the construction requirements of each leveling process, and cross-process compensation logic corrects benchmark parameters. This breaks the limitations of independent calculation of single-process parameters, comprehensively considering construction interference caused by pre-construction defects, environmental changes, and material properties. It continuously optimizes the accuracy of the leveling benchmark, improving the adaptability of wall leveling parameters from the data acquisition and algorithm calculation levels, and effectively reducing flatness errors generated during the construction phase.

[0018] Second, this invention employs a dual-layer operational architecture of edge-based on-site adjustment and offline fallback for link anomalies, coupled with a closed-loop fine-tuning algorithm and a multi-dimensional operation and maintenance service system in the middle platform. This enables flexible switching between remote cloud computing and on-site control. After receiving corrected parameters from the cloud, the edge device dynamically fine-tunes the parameters based on real-time on-site conditions to generate the final execution parameters. Construction data is stored locally and transmitted back to the cloud via two paths. Locally cached data can be used as a reference for modeling subsequent processes, while transmitted data continuously enriches the cloud sample reserves, achieving closed-loop reuse of construction data. When link communication anomalies occur, parameter calculations are completed offline using a local lightweight model. After network recovery, offline data is replenished in batches, and the collection rules and model configurations are then optimized in reverse through various digital services in the middle platform, forming a full-process data feedback and iteration mechanism. This avoids construction stagnation caused by network failures, reduces the negative impact of communication environment fluctuations on intelligent leveling operations, and steadily improves the operational reliability of the entire equipment during on-site construction.

[0019] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0021] Figure 1Flowchart for an intelligent wall surface flatness laser detection and automatic leveling system; Figure 2 This diagram illustrates the data transmission between modules of an intelligent wall surface flatness laser detection and automatic leveling system. Detailed Implementation

[0022] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0023] Reference Figure 1 One embodiment of the present invention proposes an intelligent laser detection and automatic leveling system for the flatness of decorative walls. It adopts a data processing mechanism of differentiated time-series acquisition by process, local retention of raw data through federated learning, hierarchical parameter correction at the cloud edge, and offline fallback calculation for link failures. It can rely on the entire business of digital industrial internet for serial acquisition, cloud storage modeling, parameter optimization, on-site tuning, and link maintenance. Under the condition of excessive network link latency or disconnection, it can independently complete the flatness parameter calculation and leveling control by relying on the local pre-stored model at the edge, realizing automated laser leveling and intelligent leveling operations for coarse leveling, fine leveling, and precision leveling.

[0024] The system described in this embodiment specifically includes a data acquisition and control module, a cloud-based data storage module, a parameter calculation module, an edge tuning module, and a link operation and maintenance module; The data acquisition and control module is used to collect construction data of coarse leveling, fine leveling and precision leveling processes according to specifications and mark the working conditions. It uses a multi-source feature fusion algorithm to generate a standardized construction feature dataset and sends it to the cloud storage module. The cloud-based data entry module is used to receive and encrypt the dataset through the digital industrial internet, clean it, and then store it in the cloud-based process knowledge base and federated learning sample pool according to the process. The parameter calculation module is used to retrieve data from the knowledge base and sample pool, use federated learning to train a hierarchical prediction model with a convolutional neural network architecture to predict single-process benchmark parameters, train a cross-process linkage prediction model to calculate the impact compensation value of the preceding process on the subsequent process, use a cross-order benchmark correction algorithm to correct the benchmark parameters using the impact compensation value, and generate an instruction package to be sent to the edge tuning module via the digital industrial internet. The edge tuning module is used to receive the instruction packet, and uses a field closed-loop fine-tuning algorithm to fine-tune the reference parameters based on real-time deviation. The execution parameters are generated by PID control and sent to the automatic wall leveling execution mechanism, which completes the wall leveling construction work of the corresponding process. During the construction process, the flatness, working conditions and equipment operation data collected in real time are uniformly cached to the local storage unit at the edge to form cache data. After the construction, the cache data is stored and sent back in two ways to switch the control logic of the next process. The link operation and maintenance module consists of a latency detection unit, a local model switching unit, an offline data collection unit, and a middleware interaction unit. It monitors communication link latency, and when latency is abnormal, it calls a local lightweight smoothness prediction model, using cached data for offline parameter calculations instead of cloud-based computation. After recovery, it uploads offline data, relying on the middleware platform to provide diagnostic, recommendation, and update services, and adjusts collection rules and model configurations based on feedback. Figure 2 As shown.

[0025] Specifically, the technical principle of this invention lies in constructing a layered, collaborative cloud-edge distributed operation architecture. This architecture first uses a data acquisition and control module to distinguish three leveling processes and set differentiated data acquisition frequencies. It then uses various sensors to synchronously collect multi-dimensional raw construction data, including wall flatness, substrate type, ambient temperature and humidity, mortar performance, and pre-existing defects. Finally, it relies on a multi-source feature fusion algorithm to complete the normalization, conversion, and weighted fusion of multi-dimensional heterogeneous data, forming a standardized construction feature dataset with a unified format.

[0026] The cloud-based data entry module distinguishes between on-site construction operation periods and idle periods, using different data packet transmission schemes. After receiving the dataset, it removes and cleans null values ​​and out-of-range abnormal data, and then stores it in the cloud-based process knowledge base and federated learning sample pool according to the process.

[0027] The parameter calculation module follows the federated learning data isolation rules. All the original measured data from each acquisition terminal are stored on the local hardware storage medium. Only the model gradient and process statistics are uploaded to the cloud. The module relies on the aggregated gradient data to complete the iterative training of each layer of prediction model by process. Simultaneously, a cross-process linkage prediction model is built to quantify the parameter compensation amount caused by the construction deviation of the preceding process. Then, the benchmark parameters are repeatedly verified and corrected by the cross-order benchmark correction algorithm combined with the flatness limit range of each process, and finally assembled into a control command package for distribution.

[0028] After receiving the correction command from the cloud, the edge adjustment module combines the flatness deviation obtained from real-time laser detection and performs dynamic optimization of PID parameters through a closed-loop fine-tuning algorithm on-site. It outputs execution parameters that can directly drive the automatic wall leveling actuator. The automatic wall leveling actuator consists of a servo drive unit, a mortar discharge unit, an adjustable leveling scraper unit, and a walking transmission unit. The servo drive unit receives the execution parameters and controls the coordinated action of each unit to complete the mortar spreading and leveling operations. During construction, the flatness detection data, environmental condition data, equipment operating status data, and intermediate records of parameter calculations collected in real time are uniformly cached in the local temporary storage unit at the edge, forming complete construction cache data. After the construction of a single process is completed, the full-cycle time-series data is divided into two independent links for local archiving and cloud transmission, respectively, and then the next process operation logic is automatically switched.

[0029] The link operation and maintenance module consists of four parts: a latency detection unit, a local model switching unit, an offline data collection unit, and a middleware interaction unit. The latency detection unit embeds a periodic link latency detection subroutine, which continuously captures real-time communication latency parameters of the uplink and downlink command transmission links of the digital industrial internet according to a fixed timing detection frequency. The local model switching unit is used to trigger the local lightweight model call logic and cut off the cloud parameter interaction channel when the link is abnormal. The offline data collection unit is used to collect all construction data, parameter calculation records, and fine-tuning coefficient data of the offline period after the link is restored, and then batch re-upload them to the cloud after formatting. The middleware interaction unit is used to connect with the three types of digital services of diagnosis, recommendation, and update of the digital industrial internet middleware, receive optimization ledgers, and issue configuration adjustment instructions.

[0030] Optionally, when the data acquisition and control module performs the construction data acquisition operations for coarse leveling, fine leveling, and precision leveling processes, coarse leveling completes data acquisition at a fixed low-frequency cycle of once every 5-10 seconds, fine leveling completes data acquisition at a continuous medium-frequency cycle of once every 1-2 seconds, and precision leveling completes data acquisition at a dense high-frequency cycle of more than 10 times per second. The construction data entered during the acquisition process includes wall flatness data, base surface type identification data, environmental temperature and humidity data, mortar performance data, and defect marking data of previous processes. All entered data is bound to the corresponding working condition identification content.

[0031] Specifically, before the official commencement of the renovation and construction project, technicians pre-divide the construction area based on the interior wall structure. For different substrate types, such as concrete, gypsum board, and lightweight partition wall substrates, unique identification codes are pre-entered into the data collection and control module. Simultaneously, the module's program pre-stores the corresponding data collection cycle configuration documents for each of the three leveling processes. Once on-site construction begins, the data collection and control module automatically retrieves the matching data collection cycle configuration content based on the currently executing construction process. The rough leveling stage focuses on leveling the overall wall base. Due to the thick mortar layer and the gradual change in overall wall flatness, a low-frequency acquisition mode is used. A laser flatness sensor, integrated into the scraper bracket of the automatic wall leveling actuator, moves synchronously with the walking unit. The sensor periodically measures wall points at fixed intervals. Simultaneously, it captures real-time temperature and humidity parameters, the consistency and setting properties of the current batch of mortar, and the markings of defects such as hollow areas and sandiness left over from previous base treatment stages. The fine leveling stage involves precise mortar application, significantly increasing the rate of change in local unevenness. The acquisition and control module automatically switches to a medium-frequency continuous acquisition mode, shortening the sampling interval and continuously collecting comprehensive construction information. Fine leveling, as the final step in wall forming, requires the highest precision in flatness control, capturing minute variations at the millimeter or even micrometer level. The system switches to a high-frequency dense acquisition mode to increase the number of samples per unit time. All five types of construction data entries obtained in a single collection will be bound to a working condition identifier composed of the construction area number, process type number, and work team number, providing an identifier basis for subsequent cloud-based classification and storage, data traceability, and working condition classification modeling.

[0032] For example, when the interior wall decoration project of a residential unit officially commences, the work team first carries out the rough leveling work in the master bedroom space. The data collection and control module automatically retrieves the data collection configuration file corresponding to the rough leveling and determines that the current data collection interval is within the range of 5 to 10 seconds. The equipment starts the laser rangefinder sensor to scan and collect flatness data of the preset measuring points on the master bedroom wall one by one according to the fixed interval time. At the same time that the sensor collects the flatness value, the supporting environmental sensing unit collects the temperature and relative humidity values ​​of the enclosed space inside the master bedroom in real time. The mortar detection unit simultaneously records the water retention rate, initial setting time and other performance parameters of the cement mortar that arrived on the site that day. The system retrieves the previous base surface acceptance record and enters the defect number corresponding to the crack point of the base layer. The entire set of data entries are uniformly bound to the exclusive working condition identifier of the master bedroom construction area - rough leveling - No. 1 construction team. After the rough leveling of the master bedroom is completed, the system receives a process switching trigger signal. The data acquisition and control module automatically loads the fine leveling configuration rules, switches the sampling period to a continuous acquisition mode of 1 to 2 seconds, and simultaneously increases the density of laser measuring points. It continuously collects all-dimensional construction data and updates the work condition label to "Master Bedroom Construction Area - Fine Leveling - Construction Team No. 1." Once the fine leveling process is complete, the system automatically switches the acquisition logic again to the fine leveling stage, increasing the sampling frequency to a high-frequency acquisition standard of more than ten times per second, refining the spacing of laser measuring points on the wall, and simultaneously replacing all acquired data with the corresponding work condition label for fine leveling. After the data acquisition and control module summarizes the original acquired data with work condition labels throughout the entire process, it uses the built-in multi-source feature fusion algorithm mathematical expression: in: This represents the numerical value corresponding to the wall surface flatness deviation. These are converted values ​​for the ambient temperature and humidity at the construction site. These are the collected values ​​corresponding to the physical and chemical properties of the mortar on site. This is a converted numerical value for the defect labeling of the preceding construction process. , , , The dynamic weighting coefficients are issued via the digital industrial internet. The dynamic weighting coefficients are calculated using the following formula: ; ; ; ;in: This represents the average deviation of wall flatness under historical working conditions. This is the average value calculated based on historical temperature and humidity levels under the same working conditions. The average value of physicochemical properties of mortar under the same working conditions in history. This represents the average value of defect transformation in previous processes under the same historical working conditions.

[0033] The module sequentially substitutes the measured values ​​collected in the current period and the dynamic weight coefficients sent in real time from the cloud into the formula to complete the weighted calculation. After the calculation is completed, a standardized construction feature dataset in a unified format is generated, which is then packaged and sent to the cloud storage module via the digital industrial internet.

[0034] Optionally, the process of the cloud-based data entry module receiving standardized construction feature datasets is divided into two transmission periods: during the construction operation period, the dataset is split into multiple sub-data packets and transmitted in segments with encryption; during the downtime period, the entire dataset is collected, packaged, and transmitted with encryption. The data cleaning operation of the cloud-based data entry module removes blank fields and abnormal content with values ​​exceeding the reasonable range of working conditions. The remaining content is stored in the corresponding partition of the cloud-based process knowledge base and the federated learning sample pool according to the division rules of coarse leveling, fine leveling, and precise leveling.

[0035] Specifically, the cloud-based data entry module incorporates an automatic time-of-day recognition program. Relying on the unified construction clock at the project site, it distinguishes between daytime on-duty construction periods and evening downtime / idle periods, designing differentiated data reception and transmission strategies for these two time periods. During on-site construction periods, the data acquisition and control module continuously generates standardized datasets. Upon receiving a single complete dataset, the cloud-based data entry module segments it according to the system's preset fixed single-packet byte limit, breaking it down into several smaller sub-data packets. Each sub-data packet is independently encrypted using a commercial encryption algorithm, and then received and stored in segments. This segmented transmission method avoids the instantaneous transmission of large-capacity data packets, which can consume bandwidth resources in the digital industrial internet and prevent data loss due to network congestion. During the evening downtime, when all on-site construction personnel are away and there is no leveling work, the data acquisition and control module automatically collects all standardized datasets generated but not yet uploaded to the cache for the entire day's process, integrates them into a single complete data packet, completes the encryption operation, and then transmits them in batches to the cloud-based data entry module. After the data packet is decrypted and verified to be complete and error-free, the cloud-based data entry module initiates an automated data cleaning process. The program iterates through all data within the data packet field by field, directly removing invalid data records with blank or missing key parameter fields. Simultaneously, it retrieves the system's pre-stored reasonable parameter ranges for each process condition, filtering out abnormal data entries where indicators such as flatness, ambient temperature and humidity, and mortar performance deviate from the reasonable construction range. After all cleaning work is completed, the system classifies the processes based on the condition identifiers inherent in each data entry. Coarse leveling data is stored in the cloud-based process knowledge base's dedicated coarse leveling storage partition, while fine and precise leveling data are similarly stored in their respective partitions. At the same time, a complete copy of the classified valid data is stored in the federated learning sample pool, providing a compliant sample data source for the subsequent parameter calculation module to train the federated learning model.

[0036] For example, during the daytime routine construction period of a whole-house renovation project, from 8:00 AM to 6:00 PM, the data collection and control module continuously outputs standardized construction feature datasets as the construction progresses. Each time the cloud-based data entry module receives a complete dataset, it splits it into multiple encrypted sub-packages according to a preset byte size and receives and stores them in segments. At 6:30 PM, all construction procedures on-site cease, and various on-site data collection devices automatically collect all the datasets left to be transmitted throughout the day, integrate them into a single overall data packet, encrypt them uniformly, and upload them in batches to the cloud-based data entry module during idle periods. After the cloud-based data entry module completes the data packet decryption and verification, it initiates batch data cleaning. During the screening process, the system identifies some data entries with blank or missing mortar performance fields and multiple abnormal data entries with coarse leveling flatness indicators exceeding the reasonable range. All of these abnormal data entries are removed and invalidated. All remaining compliant and valid data are classified into processes based on working condition tags. Coarse leveling classified data is stored in the coarse leveling partition of the process knowledge base, while fine leveling and precision leveling compliant data are stored in their respective knowledge base partitions. The valid data for the three types of processes are synchronously and completely backed up and stored in the federated learning sample pool for later use.

[0037] Optionally, the federated learning training process of the parameter calculation module follows a fixed data interaction specification. The original construction data collected by all data acquisition terminals is stored locally in the terminal's storage area and does not participate in network transmission. The data acquisition terminal only transmits model gradient iteration data and process statistics data to the parameter calculation module. The model gradient iteration data includes the gradient update values ​​of process parameters and feature weights for each process. The process statistics data includes single-process construction condition label data and process feature aggregation quantity data. After the parameter calculation module summarizes and aggregates the model gradient iteration data uploaded by all terminals, it matches the coarse leveling process, fine leveling process, and precise leveling process respectively, and independently completes the iterative training of the dedicated hierarchical prediction model for each process.

[0038] Specifically, the overall operation logic of federated learning here adopts a deployment mode of local raw data storage and cross-network interaction of gradient parameters. Each data acquisition terminal at the project construction site is equipped with an independent local solid-state storage hardware unit. All raw measured construction data collected by the equipment at the construction site through various sensors such as lasers, environment, and materials are permanently stored in the terminal's local storage space from the time of collection. The raw measured data will not be transmitted to the outside via the digital industrial internet, thus avoiding the risk of leakage of raw construction data at the data transmission link level. Each acquisition terminal completes a single round of model iteration calculation on the device locally based on the complete raw data stored on its own. After the single round of iteration is completed, the terminal only extracts two types of data and sends them to the parameter calculation module. The first type is model gradient iteration data, which is the gradient update change of various process parameters and data feature weights inside the hierarchical prediction model after this round of iteration training; the second type is process statistics data, including the process condition labels corresponding to the gradient data in this round and the total number of sample features actually used in this round of iteration training.

[0039] The parameter calculation module aggregates gradient data and process statistics transmitted from all terminals at all construction points in the project. Based on the process tags attached to the data, it groups and sorts the data. Coarse leveling gradient data is grouped into the first data group, while fine and precise leveling gradient data are grouped into two separate groups. These three groups are independent and do not overlap. The grouped datasets are then fed into the corresponding hierarchical prediction models for each process, independently performing iterative optimization training for the coarse, fine, and precise leveling hierarchical prediction models. Each process hierarchical prediction model employs a network architecture consisting of an input layer, a two-level feature extraction hidden layer, and an output layer. The input layer dimension and standardization... The construction feature dataset has a consistent feature dimension. The two-level hidden layers are set with 64 and 32 convolutional kernels respectively to extract multi-source construction features. The output layer outputs the benchmark parameters of the corresponding process, including three core control parameters: mortar paving thickness, squeegee travel speed, and squeegee pressing amount. During the synchronous training period of the three types of hierarchical prediction models, the parameter calculation module calls the cross-process associated construction data in the federated learning sample pool to carry out cross-process linkage prediction model training. Relying on the trained linkage model, the numerical impact of the construction error of the preceding leveling process on the benchmark parameters of the subsequent process is quantified. Finally, the impact compensation value corresponding to each process is generated and archived in a unified manner.

[0040] For example, in a whole-house renovation project, six independent construction sites are set up. Each of the six sites is equipped with a data acquisition terminal. Each of the six terminals independently stores all the original measured data collected throughout the entire process, such as wall flatness, ambient temperature and humidity, mortar parameters, and defect markings, on its local hard drive. All the original data is not uploaded to any cloud server. After each round of model iteration, the six terminals complete a round of model iteration calculations locally using their own original data. After the calculation is completed, each terminal extracts only the gradient update values ​​of the process parameters and feature weights for this round to form a gradient data package. This data package is then combined with the corresponding process condition labels and the total number of feature samples participating in the calculation for this round to form a statistical data package. Both types of data packages are encrypted and uploaded to the remote parameter calculation module. The parameter calculation module receives all data packets uploaded from six terminals. Based on the process tags embedded in the data packets, it splits and collects three independent sets of data: coarse leveling gradient set, fine leveling gradient set, and precision leveling gradient set. The three sets of data are imported into their respective hierarchical prediction models for parameter iterative optimization training. Simultaneously, it calls the associated samples of the three processes in the sample pool to train the cross-process linkage prediction model. It accurately calculates the compensation and correction amount of the coarse leveling construction deviation on the subsequent fine leveling reference parameters, and the compensation and correction amount of the fine leveling construction deviation on the subsequent precision leveling reference parameters. All calculated process impact compensation values ​​are uniformly stored in the module's built-in parameter database for later use.

[0041] Optionally, the parameter calculation module matches independent flatness value limits for each of the coarse leveling layer prediction model, fine leveling layer prediction model, and precise leveling layer prediction model. The coarse leveling layer prediction model is bound to a flatness value limit range of -2mm to +2mm, the fine leveling layer prediction model is bound to a flatness value limit range of -1mm to +1mm, and the precise leveling layer prediction model is bound to a flatness value limit range of -0.5mm to +0.5mm. After the cross-level benchmark correction algorithm completes a single parameter calculation, it immediately compares the calculated parameters with the flatness value limit range corresponding to the current construction process. When the calculated parameter value is within the flatness value limit range of the corresponding process, the current calculated parameter is retained. When the calculated parameter value exceeds the flatness value limit range of the corresponding process, the current calculated parameter is discarded, and the standardized construction feature dataset, benchmark parameters, and impact compensation values ​​are retrieved again, and the parameter calculation process of the cross-level benchmark correction algorithm is executed again. If the results of three consecutive calculations still exceed the limit range, the previous valid calculated parameter is used as the current calculated parameter, and an alarm message is generated.

[0042] Specifically, during the initial configuration phase of the three types of hierarchical prediction models, the parameter calculation module pre-assigns a unique and unchangeable flatness numerical constraint range to each model. The single-process benchmark parameters output by each model through basic training must be subject to the numerical constraints of the corresponding range. The parameter calculation module uses the mathematical formula of the cross-order benchmark correction algorithm: in: The single-process baseline parameters output by the hierarchical prediction model. This refers to the compensation value for the impact of the preceding process on the following process, output by the cross-process linkage prediction model. This refers to the feature values ​​corresponding to the standardized construction feature dataset output by the data acquisition and control module. The calculation formula is: ;in, For the first The theoretical standard reference parameters of the preceding process under a set of historical data, where N is the number of historical data sets used in the calculation. These are the actual construction parameters of the preceding process under the i-th set of historical data.

[0043] Perform baseline parameter correction calculations, and generate corrected output parameters in each round of formula calculations. Afterwards, the system immediately retrieves the flatness limit threshold corresponding to the current construction procedure to complete the numerical comparison and verification. When the corrected parameter falls within the limit range of the procedure, the system directly locks this set of parameters as the final instruction parameters to be issued; once the corrected parameter value is higher than the upper limit of the range or lower than the lower limit of the range, the system directly invalidates all calculation results of this round, and retrieves the corresponding standardized construction feature dataset, the original benchmark parameters of the layered model, and the cross-procedure impact compensation value from the cloud process knowledge base, and restarts the cross-level benchmark correction algorithm for a new round of parameter calculation; when the results of the same set of parameters in three consecutive rounds of calculation all exceed the flatness limit range of the corresponding procedure, the system terminates the repeated calculation operation, retrieves the valid parameters that have passed the range verification in the previous round as the final parameters to be used, and simultaneously pushes a parameter calculation anomaly alarm message to the project backend operation and maintenance management system to remind on-site technicians to check potential abnormal causes such as raw material performance and on-site construction environment. For example, the current process at the project site is fine leveling of the wall surface. The flatness limit range pre-bound by the fine leveling layer prediction model is -0.5mm to +0.5mm. The parameter calculation module retrieves the fine leveling benchmark parameters, the process impact compensation value corresponding to the previous fine leveling, and the current standardized feature data, and substitutes them into the cross-level benchmark correction formula to complete the first round of calculation. The correction parameter value generated in the first round of calculation exceeds the upper limit threshold of fine leveling flatness. The system directly discards the data of this round of calculation and retrieves all the supporting data to start the second cross-level benchmark correction calculation. The result of the second calculation still exceeds the limit range. The system discards the data again and starts the third parameter calculation. The parameter value after the third calculation still cannot fall into the preset value range of fine leveling. The system triggers the fallback execution logic after three consecutive exceedances. The system retrieves the valid parameters that passed the previous round of verification as the control parameters to be issued this time. At the same time, it automatically generates a fine leveling parameter continuous exceedance alarm information and pushes it to the project management backend to facilitate the management personnel to promptly investigate abnormal problems such as sudden changes in on-site mortar raw materials and environmental temperature and humidity.

[0044] Optionally, the edge adjustment module records all construction data within the complete construction cycle of a single leveling process. The complete construction cycle covers all time-series data from the start time of the process to the end time of the process. The edge adjustment module retrieves all construction data collected within the complete construction cycle and matches the flatness limit value range corresponding to the current construction process to complete the data comparison group by group. After all time-series construction data matches the flatness limit value range of the corresponding process, the edge adjustment module performs a cache storage operation on the full amount of construction data.

[0045] Specifically, the single-stage leveling process starts timing from the moment the automatic leveling equipment receives the start command until the leveling of the entire wall surface in this process is completed and the equipment receives the stop termination signal. During the entire time span of the process, all time-series flatness, environment, material, and defect marking data uploaded by the data acquisition and control module according to the corresponding process acquisition cycle are temporarily stored in the temporary cache area inside the edge adjustment module in real time. After a single leveling process is officially completed, the edge tuning module automatically locks all time-series datasets stored throughout the entire process, retrieves the pre-configured flatness limit standards for that process, and checks each set of measured flatness values ​​against the corresponding process limit range according to the time sequence of the data entries. Only when all time-series detection data for the entire process meets the flatness range constraints will the system migrate all construction data from the temporary cache to the formal cache space for archiving and storage. If some points in the time-series data have flatness values ​​that exceed the limit range, the system separately marks the abnormal point number, the value exceeding the limit, and the time of occurrence. The complete dataset with the abnormal markings is still cached and retained for subsequent cloud-based review and analysis of various causes of the exceeding problem.

[0046] For example, the fine leveling process in the secondary bedroom starts construction at a fixed time in the morning and ends at another fixed time in the morning when the fine leveling of the entire wall is completed and the machine stops. During the entire process, the data acquisition equipment continuously uploads various time-series construction data according to the dedicated medium-frequency acquisition cycle for fine leveling. The edge calibration module fully collects all data entries generated throughout the entire process. The preset flatness limit for the fine leveling process is -1mm to +1mm. After the process is completed, the module automatically verifies the actual flatness value corresponding to all time-series data one by one. After full verification, it is confirmed that all measurement point data falls within the fine leveling limit range. The system then migrates the entire set of time-series data to the formal cache partition for archiving and storage. If, during the data verification process, some wall measurement point flatness values ​​are found to exceed the fine leveling limit range, the system assigns point numbers and anomaly information labels to all exceeding measurement points. The complete dataset along with the anomaly label content is stored in the cache space for future reference.

[0047] Optionally, the construction data cached by the edge tuning module is processed independently in two paths. The first data link continuously stores the cached construction data in a dedicated local storage partition at the edge. When the next process starts data acquisition, the cached data in this local storage partition is used as historical working condition feature parameters to participate in the construction of the standardized acquisition dataset for the next process. The second data link packages the cached construction data according to the unified encapsulation format of the digital industrial internet and transmits it to the cloud storage module via the uplink encrypted channel of the digital industrial internet. After the two data links are completed, the edge tuning module calls the preset process logic configuration file to complete the switching operation from the current process to the next process control logic.

[0048] Specifically, after all construction data for a single process is cached and archived, the edge tuning module simultaneously launches two independent data processing links to operate in parallel. The first link is the local retention link: the archived construction data is stored long-term in a dedicated local storage partition on the edge device and will not be deleted from the local source files during cloud backhaul operations. When on-site construction progresses to the next leveling process and the data acquisition and control module starts a new round of on-site data acquisition, the full historical data of the previous process stored in the local partition serves as the working condition and defect feature parameters of the preceding process, and is substituted into the multi-source feature fusion algorithm to participate in the weighted fusion calculation of the standardized construction feature dataset for the new process. The second link is the cloud backhaul link: the module adjusts the internal field arrangement structure of the cached data according to the unified message encapsulation format specified by the Digital Industrial Internet, adds an encrypted verification header and data identifier field, and after encapsulation, transmits the data packet to the cloud storage module through the system's dedicated uplink encrypted channel to supplement and update the corresponding partition content of the cloud process knowledge base and the existing data in the federated learning sample pool. After all tasks on both the local storage and cloud backhaul links are completed, the edge tuning module retrieves the pre-stored process switching configuration file inside the device, automatically modifies the module's internal acquisition cycle configuration, flatness parameter reception threshold, and on-site closed-loop fine-tuning PID-related configuration parameters, and completes the switch from the current process operation logic to the next process control logic, adapting to the equipment control requirements of the next leveling process.

[0049] For example, after the completion of the entire rough leveling process in the living room area, the edge tuning module splits the full-cycle cached construction data into a dual-link processing flow. The first link permanently stores the entire set of rough leveling data in a dedicated local storage drive on the edge device. When the subsequent fine leveling process in the living room begins and the data acquisition and control module starts the fine leveling data acquisition operation, the locally retained historical rough leveling data serves as the defect characteristics and working condition parameters of the preceding process, participating in the fusion calculation of the fine leveling standardized feature dataset. The second link reorganizes the entire set of rough leveling data according to the unified specifications of the digital industrial internet, adds encrypted verification headers, and uploads it completely to the cloud storage module through the uplink encrypted channel, completing the data supplementation and update of the cloud knowledge base and sample pool. After the data storage and cloud upload tasks of both links are completed, the edge tuning module automatically reads the built-in process switching configuration document, changes the acquisition cycle from the low-frequency parameters of rough leveling to the medium-frequency parameters of fine leveling, and simultaneously replaces the internal flatness judgment threshold and PID fine-tuning configuration parameters. The entire equipment operation logic is then officially switched to adapt to the construction requirements of the fine leveling process.

[0050] Optionally, the link operation and maintenance module is configured with a fixed communication latency judgment threshold of 100ms, continuously acquires the real-time latency values ​​of the uplink and downlink of the digital industrial internet according to a fixed time sequence acquisition frequency, completes the comparison operation between the real-time latency value and the preset communication latency judgment threshold, and generates a local model call instruction and reads the lightweight flatness prediction model program file pre-stored locally at the edge when the real-time latency value is greater than 100ms or there is a continuous data interruption in the uplink and downlink.

[0051] Specifically, the link operation and maintenance module embeds a periodic link latency detection subroutine. This subroutine continuously captures real-time communication latency parameters of the uplink and downlink command transmission links in the digital industrial internet at a fixed timing frequency. The system pre-sets 100ms as the critical threshold for normal link operation. After each round of latency detection, the module compares the collected uplink and downlink real-time latency values ​​with the 100ms threshold. If either of two fault conditions occurs, the offline operation switching logic is triggered: the first fault condition is when the measured latency value of any link is greater than 100ms, and the network transmission rate cannot meet the transmission requirements for real-time control commands from the cloud and data uploads from the field; the second fault condition is when the uplink and downlink handshakes fail for an extended period, resulting in a continuous interruption of data transmission and reception. The moment the fault determination takes effect, the link maintenance module immediately sends a local model activation command to the edge tuning module, retrieving the lightweight flatness prediction model program that has been pre-installed and stored in the edge local storage hardware. This lightweight model adopts a single hidden layer fully connected network architecture, with the input dimension consistent with the real-time collected construction feature dimension. The hidden layer has 16 neurons and outputs the leveling benchmark parameters for the current process. The model weights are periodically updated by the cloud. The cloud parameter calculation interaction channel is temporarily cut off, and the system changes its operating logic to rely on the local lightweight model combined with on-site cached construction data to complete the calculation of the leveling benchmark parameters for the current process and the calculation of closed-loop fine-tuning parameters, replacing the function of the command package issued by the cloud parameter calculation module, ensuring that the automatic wall leveling actuator can continuously carry out wall leveling construction without interruption.

[0052] For example, if a signal disturbance occurs in the wireless communication environment around the project construction site, the link maintenance module's embedded detection program captures real-time uplink and downlink latency data according to a predetermined detection cycle. If the detected downlink command link's measured latency exceeds a preset 100ms threshold, the system determines that the link has entered an abnormal working state. It then generates a local model start command and sends it to the edge tuning module. The edge device reads the locally pre-stored complete lightweight flatness prediction model program, terminates the data interaction process with the remote parameter calculation module, and independently completes the flatness parameter calculation by the edge device based on the local model and the real-time collected construction data. It then directly sends control commands to the automatic wall leveling actuator. The on-site automatic leveling equipment maintains normal leveling operation throughout the process and will not experience work stoppages or waiting due to cloud communication failures.

[0053] Optionally, after the link maintenance module identifies that the digital industrial internet communication link has switched from an abnormal state to a normal connectivity state, it locks the complete timestamp interval corresponding to the abnormal working condition of the link, collects the construction data, offline status parameter calculation records, and offline working condition parameter fine-tuning coefficient data within the timestamp interval, performs unified format normalization processing on all collected data, and uploads the normalized complete data file in batches through the digital industrial internet uplink channel, with the upload target location being the cloud storage module.

[0054] Specifically, once the real-time latency of the uplink and downlink of the digital industrial internet falls below the 100ms threshold and the link handshake communication stabilizes, the link maintenance module first captures the timestamp of the link failure start time and the timestamp of the link failure recovery time, locking down all offline construction periods covered between the two timestamps. The module traverses all local storage partitions of the edge devices, completely collecting all raw construction data collected by laser sensors within the locked time interval, benchmark parameter calculation logs generated during the offline operation of the lightweight model, and PID fine-tuning coefficient ledger records actually used in the on-site closed-loop fine-tuning process. After collection, the full data is standardized by referring to the unified data format specifications of the cloud database module, and the fields are standardized, operating condition labels are completed, and anomaly annotations are improved. After standardization, the full offline data is integrated into encrypted supplementary data packets, which are then transmitted in batches to the cloud database module using the idle network bandwidth after communication recovery. After receiving the supplementary data packets, the cloud database module completes decryption and verification, merges, cleans, and archives the offline and online period data, and supplements the sample data missing from the federated learning sample pool and process knowledge base during the fault gap period.

[0055] For example, if a communication link at the construction site experiences a prolonged communication anomaly between two fixed time points on a given day, after the link maintenance module detects that the link latency has fallen back to the standard range and communication has been restored to a stable connection, the module locks the complete time interval consisting of the fault start and fault recovery. It then traverses the edge local storage partition to collect all laser-collected construction data for the fine leveling process during the fault period, all parameter records generated by the offline calculation of the local lightweight model, and archived data of various coefficients used in the on-site PID closed-loop fine-tuning process. All collected data is uniformly adjusted in field format according to the cloud database entry specifications, missing working condition labels are added, and the data is packaged and encrypted to form an offline supplementary data packet. This packet is then sent in batches to the cloud database entry module using the restored communication channel. After the cloud completes the data packet reception and decryption, it is merged into the normal construction data of the day and classified and entered into the database.

[0056] Optionally, the link operation and maintenance module calls three independent digital service units provided by the digital industrial internet platform: diagnostic service, recommendation service, and update service. The diagnostic service unit performs item-by-item verification and recording operations on the time-series transmission data, model calculation data, and process parameter data of the entire link. The recommendation service unit matches the corresponding parameter dataset in the cloud process knowledge base according to the real-time working condition tags. The update service unit generates configuration modification forms for model weight parameters and process constraint parameters. It summarizes all the ledger data generated by the operation of the three types of digital service units and revises the data collection timing rules and data annotation rules of the collection and control module, as well as the model iteration training parameters and threshold configuration parameters of the parameter calculation module, based on the ledger data.

[0057] Specifically, the link maintenance module continuously connects with three independent digital service units deployed in the digital industrial internet platform during the system's normal operation cycle: diagnostic service, recommendation service, and update service. These three service units collaborate to complete the system-wide operational data analysis and optimization plan generation. The diagnostic service unit retrieves the system's full-cycle link transmission time-series logs, the model iteration calculation logs of each round of the parameter calculation module, and the full-volume process parameter ledger executed by the edge tuning module. It checks each data transmission anomaly point, model calculation over-limit records, and process parameter anomaly items, and compiles and archives all verification results to generate a fault diagnosis ledger. The recommendation service unit captures the working condition identification tags corresponding to the currently under construction process, and uses these tags to retrieve a set of historical high-quality construction parameters with the same base material, environmental conditions, and mortar material from the massive historical construction data in the cloud-based process knowledge base. Based on the search results, it generates a recommended list of optimal construction parameters. The update service unit integrates the system defect information recorded in the diagnostic ledger with the high-quality parameter list output by the recommendation service, and compiles a configuration modification form including model internal weight parameters and flatness constraint thresholds for each process. The link operation and maintenance module summarizes all the ledger documents output by the three types of services, and remotely issues configuration change instructions based on the summarized optimized data. On the one hand, it adjusts and optimizes the collection timing parameters of each process and the specifications of mandatory fields for construction data annotation within the collection and control module. On the other hand, it simultaneously revises the configuration contents such as the iteration step size of each layer model and the flatness threshold of each process in the parameter calculation module, and realizes the periodic autonomous optimization of system configuration based on actual on-site construction data.

[0058] For example, during the process of verifying all data item by item, the diagnostic service unit of the digital industrial internet platform screened out a large number of fault records of historical coarse leveling data that failed to be entered into the cloud database due to missing working condition labeling fields, and compiled them into the diagnostic ledger. The recommendation service unit, based on similar construction working condition tags, retrieved multiple sets of high-quality weight coefficient parameters suitable for the on-site working conditions from the process knowledge base and generated a parameter recommendation list. The update service unit, combined with the contents of the first two ledgers, compiled a form for modifying collection rules and model parameters. After the link operation and maintenance module summarized all the ledger contents formed by the three services, it remotely issued a configuration change command to supplement and improve the mandatory fields of coarse leveling working condition labeling in the collection and control module. At the same time, it fine-tuned the iteration coefficients and flatness constraint thresholds of the coarse leveling layered prediction model inside the parameter calculation module, completing the configuration iteration optimization of the entire system.

[0059] Optionally, the mathematical expression of the on-site closed-loop fine-tuning algorithm used by the edge tuning module is: in: For the first The execution parameters are calculated by PID control and output to the actuator after each control cycle. This is the proportional gain coefficient, used to provide an immediate response to deviations at the current moment; This is the integral time coefficient, used to eliminate system steady-state error; The differential time coefficient is used to suppress overshoot and oscillation; The sampling period for on-site closed-loop control; For the first In each control cycle, the difference between the real-time wall flatness deviation obtained by laser detection and the flatness target value required by the current process; The deviation value of the previous control cycle; summation term This is the cumulative sum of deviations from the start of control to the current time, used for integration. , , The calculation formula is: ; ; In the formula This is the initial reference value for the scaling factor. This represents the largest flatness deviation in the history of this process. This represents the lowest flatness deviation in the history of this process. This is the initial time constant for integration. is the initial time constant for the differential.

[0060] Specifically, after the edge adjustment module receives the process benchmark correction parameters issued by the digital industrial internet from the parameter calculation module, the laser flatness detection equipment continuously scans the points on the wall surface to be leveled in real time to obtain the measured flatness values. The module then calculates the difference between the measured flatness values ​​and the preset flatness target values ​​for this process to obtain the current deviation. The parameters are then fully substituted into the PID calculation formula for the closed-loop fine-tuning algorithm. The formula includes the proportional gain coefficient. The corresponding proportional term can quickly respond to instantaneous flatness deviations and adjust the actuator's movement amplitude in real time; integral time coefficient Matching sampling period The sum of the total deviations over the entire cycle forms the integral term, which relies on the accumulated historical deviation values ​​to eliminate the static leveling error caused by long-term equipment operation; the differential time coefficient Combining the differential term formed by the difference in deviation between adjacent cycles and the sampling period, the overshoot and mechanical oscillation caused by drastic changes in the amplitude of the leveling actuator are suppressed by relying on the rate of change of deviation; the three calculation results are superimposed and summed to generate the final execution parameters for the current control cycle. The generated execution parameters directly drive the servo drive unit of the automatic wall leveling actuator, which in turn controls the discharge volume of the mortar discharge unit and the pressing height and travel speed of the leveling scraper unit, adjusting the mortar spreading thickness to achieve closed-loop dynamic leveling of the wall surface. After a single cycle of calculation, the system saves the current deviation. The deviation e(n-1) of the previous cycle required for the next control cycle calculation is used to continuously complete the closed-loop fine-tuning operation of the entire process.

[0061] For example, during the fine leveling construction phase of the kitchen and bathroom space, the edge alignment module receives fine leveling benchmark correction parameters from the cloud, and the laser sensor continuously collects measured flatness data of the wall surface according to the high-frequency acquisition cycle of fine leveling. Within each closed-loop control cycle, the system calculates the difference between the measured flatness and the target flatness of the process to obtain the deviation for the current period. Set the preset proportional gain coefficient Integral time coefficient Differential time coefficient Closed-loop sampling period Current period deviation Substituting the previous cycle's retained deviation e(n-1) and the total cycle deviation summation term into the closed-loop fine-tuning calculation formula, and performing the proportional, integral, and differential terms item by item, the final execution parameters are obtained by summing them up. The automatic wall leveling actuator is controlled by this parameter to adjust the material output rate, scraper pressure and travel speed in real time. As construction progresses, the deviation values ​​of each period are continuously saved for the next cycle calculation. The entire process relies on a closed-loop algorithm to dynamically eliminate wall flatness deviations, ensuring that the precision of kitchen and bathroom wall leveling construction meets the preset limit standards.

[0062] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A smart laser detection and automatic leveling system for the flatness of decorative walls, characterized in that, It includes a data acquisition and control module, a cloud-based data storage module, a parameter calculation module, an edge tuning module, and a data link maintenance module. The data acquisition and control module is used to collect construction data of coarse leveling, fine leveling and precision leveling processes according to specifications and mark the working conditions. It uses a multi-source feature fusion algorithm to generate a standardized construction feature dataset and sends it to the cloud storage module. The cloud-based data entry module is used to receive and encrypt the dataset through the digital industrial internet, clean it, and then store it in the cloud-based process knowledge base and federated learning sample pool according to the process. The parameter calculation module is used to retrieve data from the knowledge base and sample pool, use federated learning to train a hierarchical prediction model to predict single-process benchmark parameters, train a cross-process linkage prediction model to calculate the impact compensation value of the preceding process on the following process, and use a cross-order benchmark correction algorithm to correct the benchmark parameters using the impact compensation value, and generate an instruction package to be sent to the edge tuning module via the digital industrial internet. The edge tuning module is used to receive the instruction packet, and uses a field closed-loop fine-tuning algorithm to fine-tune the reference parameters based on the real-time deviation. The execution parameters are generated by PID control and sent to the automatic wall leveling execution mechanism. The automatic wall leveling execution mechanism completes the wall leveling construction operation of the corresponding process according to the execution parameters. During the construction process, the flatness, working conditions and equipment operation data collected in real time are uniformly cached to the local storage unit at the edge to form cache data. After the construction, the cache data is stored and sent back in two ways to switch the control logic of the next process. The link operation and maintenance module is used to monitor communication link latency. When latency is abnormal, it calls the local lightweight flatness prediction model and uses cached data for offline parameter calculation instead of cloud calculation. After recovery, it uploads offline data and provides diagnostic, recommendation and update services based on the middle platform. The collection rules and model configuration are adjusted according to the feedback.

2. The intelligent wall surface flatness laser detection and automatic leveling system according to claim 1, characterized in that, When the data acquisition and control module performs the construction data acquisition operations for coarse leveling, fine leveling, and precision leveling processes, coarse leveling completes data acquisition at a fixed low-frequency cycle of once every 5-10 seconds, fine leveling completes data acquisition at a continuous medium-frequency cycle of once every 1-2 seconds, and precision leveling completes data acquisition at a dense high-frequency cycle of more than 10 times per second. The construction data entered during the acquisition process includes wall flatness data, base surface type identification data, environmental temperature and humidity data, mortar performance data, and defect marking data of previous processes. All entered data is bound to the corresponding working condition identification content.

3. The intelligent wall surface flatness laser detection and automatic leveling system according to claim 1, characterized in that, The process of receiving standardized construction feature datasets by the cloud-based data entry module is divided into two transmission periods: during the construction operation period, the dataset is split into multiple sub-data packets and transmitted in segments with encryption; during the downtime period, the entire dataset is collected, packaged, and transmitted with encryption. The data cleaning operation of the cloud-based data entry module removes blank fields and abnormal content with values ​​exceeding the reasonable range of working conditions. The remaining content is stored in the corresponding partition of the cloud-based process knowledge base and the federated learning sample pool according to the division rules of coarse leveling, fine leveling, and precise leveling.

4. The intelligent wall surface flatness laser detection and automatic leveling system according to claim 1, characterized in that, The federated learning training process of the parameter calculation module follows a fixed data interaction specification. The original construction data collected by all data acquisition terminals is stored locally on the terminal and does not participate in network transmission. The data acquisition terminals only transmit model gradient iteration data and process statistics data to the parameter calculation module. The model gradient iteration data includes the gradient update values ​​of process parameters and feature weights for each process. The process statistics data includes the single-process construction condition label data and the number of process feature collections. After the parameter calculation module summarizes and collects the model gradient iteration data uploaded by all terminals, it matches the coarse leveling process, fine leveling process, and precision leveling process respectively, and independently completes the iterative training of the dedicated hierarchical prediction model for each process.

5. The intelligent wall surface flatness laser detection and automatic leveling system according to claim 1, characterized in that, The parameter calculation module matches independent flatness value limits for the coarse leveling layer prediction model, the fine leveling layer prediction model, and the precision leveling layer prediction model. The coarse leveling layer prediction model is bound to a flatness value limit range of -2mm to +2mm, the fine leveling layer prediction model is bound to a flatness value limit range of -1mm to +1mm, and the precision leveling layer prediction model is bound to a flatness value limit range of -0.5mm to +0.5mm. After completing a single parameter calculation, the cross-level benchmark correction algorithm immediately compares the corrected calculation parameters with the flatness value limit range of the current construction process. If the corrected calculation parameter value is within the flatness value limit range of the corresponding process, the current corrected calculation parameter is retained. If the corrected calculation parameter value exceeds the flatness value limit range of the corresponding process, the current corrected calculation parameter is discarded, and the standardized construction feature dataset, benchmark parameters, and impact compensation values ​​are retrieved again, and the parameter calculation process of the cross-level benchmark correction algorithm is executed again. If the result of n consecutive calculations still exceeds the limit range, the valid calculation parameter that passed the previous verification in the current process is used as the current calculation parameter, and an alarm message is generated.

6. The intelligent wall surface flatness laser detection and automatic leveling system according to claim 1, characterized in that, The edge adjustment module records all construction data within the complete construction cycle of a single leveling process. The complete construction cycle covers all time-series data from the start time to the end time of the process. The edge adjustment module retrieves all construction data collected within the complete construction cycle and matches the flatness limit value range corresponding to the current construction process to complete the data comparison group by group. After all time-series construction data matches the corresponding flatness limit value range of the process, the edge adjustment module performs a cache storage operation on the full amount of construction data.

7. The intelligent wall surface flatness laser detection and automatic leveling system according to claim 1, characterized in that, The construction data cached by the edge tuning module is processed independently in two paths. The first data link continuously stores the cached construction data in a dedicated local storage partition at the edge. When the next process starts data acquisition, the cached data in the dedicated local storage partition at the edge is used as historical working condition feature parameters to participate in the construction of the standardized acquisition dataset for the next process. The second data link packages the cached construction data according to the unified encapsulation format of the digital industrial internet and transmits it to the cloud storage module via the uplink encrypted channel of the digital industrial internet. After the two data links are completed, the edge tuning module calls the preset process logic configuration file to complete the switching operation from the current process to the next process control logic.

8. The intelligent wall surface flatness laser detection and automatic leveling system according to claim 1, characterized in that, The link operation and maintenance module is configured with a fixed communication latency judgment threshold of 100ms. It continuously acquires the real-time latency values ​​of the uplink and downlink of the digital industrial internet at a fixed time sequence acquisition frequency, and completes the comparison operation between the real-time latency value and the preset communication latency judgment threshold. When the real-time latency value is greater than 100ms or there is a continuous data interruption in the uplink and downlink, it generates a local model call instruction and reads the lightweight flatness prediction model program file pre-stored locally at the edge.

9. The intelligent wall surface flatness laser detection and automatic leveling system according to claim 1, characterized in that, After the link maintenance module detects that the digital industrial internet communication link has switched from an abnormal state to a normal connectivity state, it locks the complete timestamp interval corresponding to the abnormal working condition of the link, collects the construction data, offline status parameter calculation records, and offline working condition parameter fine-tuning coefficient data within the timestamp interval, performs unified format normalization processing on all collected data, and uploads the normalized complete data file in batches through the digital industrial internet uplink channel. The upload target location is the cloud storage module.

10. The intelligent wall surface flatness laser detection and automatic leveling system according to claim 1, characterized in that, The link operation and maintenance module calls three independent digital service units provided by the digital industrial internet platform: diagnostic service, recommendation service, and update service. The diagnostic service unit performs item-by-item verification and recording operations on the time-series transmission data, model calculation data, and process parameter data of the entire link. The recommendation service unit matches the corresponding parameter dataset in the cloud process knowledge base according to the real-time working condition tags. The update service unit generates configuration modification forms for model weight parameters and process constraint parameters. It summarizes all the ledger data generated by the operation of the three types of digital service units. The link operation and maintenance module revises the data acquisition timing rules and data annotation rules of the acquisition and control module, as well as the model iteration training parameters and threshold configuration parameters of the parameter calculation module, based on the ledger data.

Images