A building whole life cycle wisdom management platform

Abstract

This invention discloses a building lifecycle intelligent maintenance management platform, relating to the field of intelligent maintenance management technology. The platform includes: a building design management unit that generates a design BIM model carrying design attributes and equipment information; a building construction management unit that iteratively generates a construction BIM model based on the design model; an as-built BIM model obtained by associating construction data; and an association mapping module in the intelligent operation and maintenance management unit that establishes a mapping relationship between the two models and analyzes equipment operation data and fault records using a hybrid prediction model, outputting health scores and remaining usable life predictions for each equipment component. This invention generates predictive maintenance work orders and determines execution priorities when the health score or remaining usable life prediction result of any equipment component in the maintenance decision module meets preset maintenance trigger conditions, thereby avoiding biased judgments and inaccurate control of maintenance timing, and ultimately improving the accuracy of building equipment operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of intelligent maintenance management technology, specifically to an intelligent maintenance management platform for the entire life cycle of a building. Background Technology

[0002] The entire building lifecycle typically encompasses the design, construction, and operation and maintenance phases. The types of information generated and used at each phase are diverse and come from a wide range of sources.

[0003] Building Information Modeling (BIM) technology is widely used in the design phase, enabling multidisciplinary collaborative design and clash detection, and generating digital models carrying design attributes and spatial information. In the construction phase, IoT sensor technology is gradually introduced, enabling real-time data collection on the construction site environment, key process quality, and equipment installation, providing data support for construction process control. In the operation and maintenance phase, some buildings have begun to combine BIM with IoT sensors to display equipment operating status through 3D visualization, assisting property management personnel in daily inspections and troubleshooting.

[0004] The limitations of existing technologies include at least the following problems: existing BIM models are only independent static models for each stage, making it difficult to achieve orderly iteration and complete data inheritance from design models to construction models and then to operation and maintenance models. This makes it difficult to accurately associate and accumulate construction data in the construction stage into model components, and difficult to rely on a unified digital model carrier to combine historical equipment data and real-time sensor data to carry out quantitative assessment of equipment health and life prediction in the operation and maintenance stage. As a result, it is difficult to quantitatively assess the health status of equipment and accurately predict its remaining usable life, and consequently, it is difficult to scientifically determine the execution priority of maintenance tasks for multiple pieces of equipment to be maintained, thus limiting the level of refinement and intelligence of digital operation and maintenance of buildings. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a building lifecycle intelligent maintenance management platform, which solves the problem that existing technologies struggle to integrate data throughout the entire lifecycle, thus hindering predictive maintenance and intelligent scheduling of equipment.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a building lifecycle intelligent maintenance management platform, comprising: a building design management unit for generating a design BIM model carrying design attributes and equipment information; a building construction management unit for generating a construction BIM model based on the design BIM model, and receiving construction data collected by IoT sensors during the construction phase, associating it with the construction BIM model to form an as-built BIM model; an intelligent operation and maintenance management unit, including an association mapping module for converting the as-built BIM model into an operation and maintenance BIM model, and establishing a data mapping relationship between each equipment component in the operation and maintenance BIM model and its corresponding IoT sensor; a hybrid prediction and evaluation module for acquiring historical operating data, historical fault records, and real-time operating status parameters of each equipment component, constructing a hybrid prediction model, and outputting the health score and remaining usable life prediction value of each equipment component; and a maintenance decision module for generating a predictive maintenance work order and determining the execution priority when the health score or remaining usable life prediction result of any equipment component meets the preset maintenance triggering conditions.

[0007] Furthermore, the specific steps for forming the as-built BIM model are as follows: receive construction data collected by IoT sensors during the construction phase, and perform time-series alignment and outlier cleaning; extract the unique identifier code of the construction BIM model, and establish an association mapping between the cleaned construction data and the construction BIM model according to the unique identifier code and collection timestamp; write the construction data into the construction BIM model in the form of an extended attribute set to generate the as-built BIM model.

[0008] Furthermore, the specific steps for converting the as-built BIM model into an operation and maintenance BIM model are as follows: perform attribute cleaning on the as-built BIM model; complete the operation and maintenance information on the attribute-cleaned as-built BIM model; and perform lightweight processing on the completed as-built BIM model to obtain the operation and maintenance BIM model.

[0009] Furthermore, the specific steps for establishing the data mapping relationship between each equipment component and the corresponding IoT sensor in the operation and maintenance BIM model are as follows: extract the operation and maintenance asset code and spatial coordinates of each equipment component in the operation and maintenance BIM model; obtain the equipment identification and installation location information of each IoT sensor deployed in the building; and establish a one-to-one or one-to-many mapping relationship between each equipment component and the corresponding IoT sensor, using the operation and maintenance asset code as the first matching key and the spatial coordinates as the second matching key.

[0010] Furthermore, the hybrid prediction model is composed of a physical mechanism model and a data-driven model, and the specific steps for outputting the health score and remaining usable life prediction value of each equipment component are as follows: The physical mechanism model is based on the equipment operation mechanism and uses real-time operating status parameters to simulate the degradation behavior of the equipment, and outputs a first health score and a first remaining usable life; the data-driven model is trained using historical operating data and historical fault records, inputs real-time operating status parameters, and outputs a second health score and a second remaining usable life; the hybrid prediction model performs weighted fusion of the first health score and the second health score, and the first remaining usable life and the second remaining usable life, respectively, and outputs the health score and remaining usable life prediction value of each equipment component.

[0011] Furthermore, the training steps of the data-driven model are as follows: Data cleaning is performed on the historical operational data; the cleaned historical operational data is sliced ​​by time window, and multi-dimensional features are extracted within each time window, including time-domain statistical features, frequency-domain features, and time-series trend features; the feature vector is used as input, and the fault type corresponding to the time window is used as the classification label; the classification error is calculated using cross-entropy loss, and the regression error is calculated using mean squared error loss. The two are weighted and summed as the total loss function. The network parameters are iteratively optimized through the backpropagation algorithm until the total loss function converges, thus obtaining the data-driven model.

[0012] Furthermore, the preset maintenance triggering conditions adopt a dual judgment logic: when the health score is lower than the preset dynamic threshold, it is determined that maintenance needs to be triggered; when the remaining usable life prediction value is less than the preset maintenance window duration, it is determined that maintenance needs to be triggered; when either judgment logic is satisfied, a predictive maintenance work order is generated.

[0013] Further, the specific steps for determining the execution priority are as follows: standardize the health score and the remaining usable life prediction value; comprehensively process the standardized health score and the remaining usable life prediction value to generate a maintenance priority score value, and sort them in descending order to generate a maintenance priority table; determine the execution priority based on the maintenance priority table.

[0014] The present invention has the following beneficial effects:

[0015] This intelligent maintenance management platform for the entire building lifecycle generates a construction BIM model through iterative generation based on the design BIM model. After cleaning and accurately associating various sensor data collected during the construction phase with the corresponding components in the model, a complete as-built BIM model is formed. This enables the orderly retention and unified inheritance of data throughout the entire building construction process. By standardizing, completing, and lightweighting the as-built model, it adapts to the needs of operation and maintenance scenarios. Combining asset coding and spatial coordinate matching, it establishes a precise correspondence between equipment components and sensor data, ensuring that equipment operation data can be retrieved synchronously in real time. At the same time, through a weighted fusion of the two models, it obtains health status and lifespan results that more closely reflect the actual working conditions of the equipment. Maintenance tasks are triggered based on both health and remaining lifespan dimensions, and maintenance priority scores are calculated and sorted to determine the work sequence. This avoids problems such as biased judgment and inaccurate control of maintenance timing, thereby improving the accuracy of building equipment operation and maintenance and the overall management efficiency.

[0016] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0017] Figure 1 This is a block diagram of a building lifecycle intelligent maintenance management platform according to the present invention.

[0018] Figure 2 This is a flowchart illustrating the specific steps involved in outputting the health score and remaining usable life prediction value of each equipment component in a building lifecycle intelligent maintenance management platform according to the present invention. Detailed Implementation

[0019] Please see Figure 1 The present invention provides a technical solution: a building life cycle intelligent maintenance management platform, including: a building design management unit, used to generate a design BIM model carrying design attributes and equipment information, and to transfer the design BIM model to the construction stage through a unified data standard;

[0020] The building construction management unit is used to generate a construction BIM model based on the design BIM model, and to receive construction data (including construction environment data, key process quality data and equipment arrival and installation data) collected by IoT sensors during the construction phase. It associates the above data with the corresponding components of the construction BIM model to form an as-built BIM model carrying complete construction process information.

[0021] The intelligent operation and maintenance management unit includes an association mapping module, which is used to convert the as-built BIM model into an operation and maintenance BIM model after attribute cleaning, operation and maintenance information completion and lightweight processing, and to establish the data mapping relationship between each equipment component in the operation and maintenance BIM model and the corresponding IoT sensor, so that the equipment components can reflect the real-time operating status synchronously.

[0022] The hybrid prediction and evaluation module is used to acquire historical operating data, historical fault records and real-time operating status parameters of each equipment component, build a hybrid prediction model, and output the health score and remaining usable life prediction value of each equipment component.

[0023] The maintenance decision module is used to generate predictive maintenance work orders and determine execution priorities when the health score or remaining usable life prediction result of any equipment component meets the preset maintenance trigger conditions.

[0024] The specific steps for generating a design BIM model that carries design attributes and equipment information are as follows:

[0025] Obtain BIM models for each discipline in architectural design. Each discipline's BIM model includes at least an architectural model, a structural model, and a mechanical and electrical (MEP) model. Specifically, this involves exporting the original BIM models for each discipline from architectural design software, structural design software, and MEP design software, converting them into the IFC standard format, importing them into the platform, and aligning the coordinate systems and unifying the elevations of each discipline's models.

[0026] The architectural model includes components such as walls, floors, doors, and windows; the structural model includes load-bearing components such as beams, columns, and foundations; and the mechanical and electrical model includes HVAC equipment, water supply and drainage pipes, and power distribution equipment.

[0027] Multi-disciplinary collaborative clash detection is performed on the BIM models of various disciplines. After marking and correcting component conflicts, an integrated BIM model is generated. The specific steps are as follows:

[0028] The various professional models are superimposed in the same 3D space. The bounding box hierarchical tree algorithm is used to detect hard and soft collisions. Hard collisions refer to direct intersections of components, while soft collisions refer to the distance between components being less than a preset minimum operating space threshold. The detection results are displayed in a list, including the collision component number, type, coordinates, and screenshot.

[0029] For common collision types, the system recommends adjustment schemes and performs semi-automatic corrections according to preset avoidance rules; for those that cannot be handled automatically, designers make manual adjustments, and after correction, they are merged into an integrated BIM model, retaining the original attributes and professional classifications of each component;

[0030] In the integrated BIM model, design attribute parameters are attached to each component. These design attribute parameters include at least the component's geometric dimensions, material grade, fire rating, and design service life, specifically:

[0031] The corresponding field list is read from the preset design attribute template according to the component type, and the attribute values ​​are automatically extracted from the original professional model. If there is no data, it is manually entered.

[0032] The structural column connection includes cross-sectional dimensions, concrete strength grade, seismic resistance grade, and fire resistance grade; the building partition wall connection includes wall thickness, block material grade, and fire resistance grade; the electromechanical pipeline connection includes pipe diameter, wall thickness, material, and insulation material thickness. The connected attributes are stored as an extended attribute set.

[0033] For electromechanical equipment components in the integrated BIM model, attach equipment information parameters. The equipment information parameters should include at least the equipment name, equipment model, rated power, rated voltage, and design operating condition parameters. Specifically, attach equipment information parameters for electromechanical equipment components such as chillers, cooling towers, water pumps, air handling units, distribution cabinets, transformers, and elevator traction machines.

[0034] Equipment names should be formatted according to standard specifications. Rated power, voltage, cooling capacity, etc., should be imported from the equipment selection table. Design operating parameters should include supply and return water temperatures, outdoor temperature and humidity, and design load. Parameters should be automatically matched and filled from the equipment library; if no corresponding model is found, they should be manually entered.

[0035] The integrated BIM model after the connection is completed will be output as a design BIM model carrying design attributes and equipment information, specifically as follows:

[0036] The integrated BIM model, after collision correction and attribute attachment, is exported to the platform's internal data format and a lightweight visualization file is generated simultaneously. This design BIM model serves as the starting point for the entire lifecycle data flow and is transmitted to the construction phase management unit through a unified data interface.

[0037] Specifically, the steps to create an as-built BIM model are as follows:

[0038] The system receives construction data collected by IoT sensors during the construction phase. This data includes at least construction environment data, key process quality data, and equipment arrival and installation data. The system then performs time-series alignment and outlier cleanup on the construction data, specifically:

[0039] The IoT sensors used during the construction phase include: environmental monitoring instruments deployed in various areas of the construction site to collect construction environment data such as temperature, humidity, inhalable particulate matter concentration, and noise; temperature sensors installed in the concrete curing area to collect key process quality data such as temperature changes during the curing period after concrete pouring; and handheld equipment entry scanning terminals to scan the equipment's factory QR code to collect equipment entry and installation data such as equipment arrival time, actual installation location coordinates, and installation personnel information.

[0040] The aforementioned sensors upload data to the platform via 4G or Wi-Fi. After receiving the data, the platform aligns the data from different sources according to timestamps and uses sliding window interpolation to solve the time matching problem of different acquisition frequencies. Values ​​that exceed the physical possible range are marked as abnormal and removed. Short-term missing data are filled by linear interpolation or mean before and after. The cleaned data is temporarily stored in the construction data cache library according to data type.

[0041] Extract the unique identifier code of each component in the construction BIM model, and establish an association mapping between the cleaned construction data and the corresponding components in the construction BIM model according to the unique identifier code and the collection timestamp. Specifically:

[0042] When the construction BIM model is generated, a globally unique identifier code is assigned to each component. The platform parses the construction BIM model, extracts the unique identifier codes of all components and their spatial locations, and automatically associates the construction environment data with the corresponding components based on the inclusion relationship between the sensor deployment location coordinates and the spatial range of the components.

[0043] For key process quality data, a mapping is established between component numbers and unique identifier codes in the construction records; for equipment arrival and installation data, it is precisely matched with the unique identifier codes of the equipment components that have been attached in the construction BIM model, and the association mapping results form a triplet record table of timestamp, component code and data value.

[0044] The construction data associated with each component is written into the construction BIM model in the form of an extended attribute set, generating an as-built BIM model carrying construction process information, specifically as follows:

[0045] In the construction BIM model, a construction extended attribute field group is created for each component. The triplet records are grouped according to the component code. Key statistical information is extracted from each group of data: environmental data takes the mean and maximum values ​​during construction; key process quality data retains the measured values ​​and qualification judgment results at key time points; equipment arrival and installation data records the actual installation time, installation personnel, commissioning and operation parameters, and final installation coordinates. The statistical information and the storage index of the original time series data are written into the construction extended attribute set of the corresponding component, and the construction BIM model is updated to the as-built BIM model.

[0046] The specific steps to convert the as-built BIM model into an operation and maintenance BIM model are as follows:

[0047] The as-built BIM model undergoes attribute cleaning, removing information on temporary facilities during the construction phase, construction scaffolding, and construction process markers unrelated to permanent operation and maintenance. Specifically:

[0048] The platform traverses all components of the as-built BIM model, filters them according to component classification tags, marks components classified as temporary facilities (temporary supports, formwork, construction elevator wall attachments, etc.) as non-maintenance objects and hides them from the maintenance view; construction scaffolding information is completely removed, and construction process marking data unrelated to permanent maintenance (concrete pouring batch number, construction team record, temporary measurement point marking, etc.) is filtered through an attribute field name whitelist mechanism, retaining design attributes, equipment information and key construction quality data, and eliminating pure construction management process markings;

[0049] After attribute cleaning, the as-built BIM model is supplemented with operation and maintenance information, and operation and maintenance asset codes, standard maintenance cycles, warranty expiration dates, and supplier information are attached to each equipment component. Specifically:

[0050] For each equipment component, the maintenance asset code is matched from the enterprise asset management system or equipment ledger database based on its equipment name and model. If no match is found, the code is automatically generated according to the coding rules (building number plus system number plus equipment serial number).

[0051] The standard maintenance cycle is read from the standard maintenance outline library according to the equipment type; the warranty expiration date is calculated based on the installation date plus the contract warranty period; supplier information is imported from the procurement system or contract, and the completed information is linked to the operation and maintenance attribute set of the equipment components;

[0052] The completed as-built BIM model is then lightweighted by reducing the amount of model data while preserving the spatial topological relationships of equipment components, resulting in an operation and maintenance BIM model, specifically as follows:

[0053] Lightweight processing employs geometric simplification and instantiation techniques. For building components that are not critical to operation and maintenance, such as walls and floors, bounding box approximations or low-face-count meshes are used to replace the original high-precision geometry, with the geometric simplification error controlled within two centimeters.

[0054] For equipment components, retain their external features but remove internal details. For example, for water pumps, retain the outer shell outline but delete the models of internal parts such as impellers and bearings. At the same time, for components of the same model that appear repeatedly, such as lamps and air vents of the same model, instantiate them and store only one geometric prototype, which is then rendered in the scene by reference.

[0055] Spatial topology relationships, such as the space to which equipment belongs, pipeline connection relationships, and electrical upstream and downstream relationships, are preserved through a topology diagram. The processed model is the operation and maintenance BIM model.

[0056] The specific steps for establishing the data mapping relationship between each equipment component and its corresponding IoT sensor in the operation and maintenance BIM model are as follows:

[0057] Extract the operation and maintenance asset codes and spatial coordinates of each equipment component in the operation and maintenance BIM model, specifically as follows:

[0058] Traverse the BIM model of operation and maintenance, filter out all equipment components, read the operation and maintenance asset code from the operation and maintenance attribute set, extract the positioning point coordinates from the geometric definition (take the center of the bottom surface of the equipment enclosure box or the installation base point), and pair the asset code with the spatial coordinates to generate a list of equipment component information.

[0059] Obtain the device identification and installation location information of each IoT sensor deployed within the building, specifically:

[0060] Through the interface with the IoT data platform, basic information of all registered sensors is obtained, including the sensor's unique device identifier, sensor type (such as temperature, vibration, current, pressure, etc.), description of the measured object (such as the outlet water temperature of chiller unit CH-01), and installation location coordinates. The installation location information is recorded by scanning a code during the sensor installation and acceptance phase of construction, including the floor, room number, and precise coordinates. The sensor information list is indexed using the device identifier as the key, awaiting the establishment of a mapping with the device components.

[0061] Using the maintenance asset code as the first matching key and the spatial coordinates as the second matching key, a one-to-one or one-to-many mapping relationship is established between each device component and its corresponding IoT sensor, specifically as follows:

[0062] Perform the first level of matching: Traverse the list of device component information. For each device component, search the IoT sensor information list for its operation and maintenance asset code to see if there is a sensor whose device identification number or sensor description field contains the asset code. For example, if the sensor device identification number contains the asset code CH-01, a direct match can be established.

[0063] If the first-level matching is successful, the corresponding sensor number is bound to the equipment component code. If the first-level matching fails, for example, if the sensor is not directly labeled with an asset code, the second-level matching is performed: the spatial coordinates of the equipment component and the installation location coordinates of the sensor are calculated to find the sensor that is closest to the device and located on the same floor and in the same room. Additionally, reasonable pairing of sensor type and possible monitoring parameters of the device is considered.

[0064] After matching is complete, a mapping table is generated, with each device component corresponding to one or more sensors. The mapping relationship is stored in the platform's mapping database and supports dynamic updates. When real-time sensor data enters the platform, the corresponding device component is automatically found through the mapping relationship, and its operating status display in the 3D view is updated.

[0065] In this implementation plan, after standardization, the data is uniformly associated with the corresponding components of the model, achieving complete retention and orderly inheritance of information throughout the entire construction process. By cleaning up redundant content in the model and supplementing it with various equipment data required for operation and maintenance, the model can be adapted to the actual usage scenarios of subsequent operation and maintenance work. At the same time, the model content is simplified through a lightweight approach, ensuring normal display and operation of the model while retaining key structural relationships. Furthermore, based on dual matching rules, a corresponding connection between equipment components and sensing devices is established, realizing automatic linkage and synchronization between on-site monitoring data and the digital model. This allows operation and maintenance work to be carried out based on complete and accurate full-cycle building data, thereby improving the standardization of building operation and maintenance management.

[0066] Specifically, such as Figure 2 As shown, the hybrid prediction model is composed of a physical mechanism model of the equipment and a data-driven model. The specific steps for outputting the health score and remaining usable life prediction value of each equipment component are as follows:

[0067] The physical mechanism model is based on the equipment's operating mechanism and uses real-time operating status parameters to simulate the equipment's degradation behavior, outputting a first health score and a first remaining usable life, specifically as follows:

[0068] Physical mechanism models are constructed separately for different equipment types, taking centrifugal chillers as an example:

[0069] The platform uses equipment design parameters (rated cooling capacity, coefficient of performance, heat exchange area, etc.) and real-time operating status parameters (evaporation temperature, condensation temperature, compressor current, lubricating oil temperature, etc.) as inputs, and calculates the current actual coefficient of performance and volumetric efficiency through thermodynamic and heat transfer models.

[0070] An exponential degradation model is used to describe the performance degradation trend. The degradation rate is determined by factors such as the ratio of current to rated current and the lubricating oil quality decay rate. Based on the degree of deviation between the current operating state and the design conditions, the first health score (out of 100, with the full score being completely healthy) is calculated. Based on the degradation trend, the remaining time to reach the failure threshold (such as a 30% decrease in performance coefficient) is predicted, which is the first remaining usable life. Similarly, degradation simulation models are constructed for other equipment such as water pump units and transformers based on their respective operating mechanisms.

[0071] The data-driven model is trained using historical operational data and historical fault records. It takes real-time operational status parameters as input and outputs a second health score and a second remaining usable lifespan, specifically:

[0072] The data-driven model is a deep neural network. After training, it is deployed on the model inference engine. It receives real-time operating status parameters of the current time window, extracts features, and inputs them into the network. It outputs the probability vector of the device in each fault category and the remaining usable life value. The information entropy of the fault probability vector is converted into a second health score (the higher the probability of no fault, the closer the score is to the full score). The second remaining usable life is directly taken from the network regression output.

[0073] The hybrid prediction model weights and fuses the first health score and the second health score, as well as the first remaining usable life and the second remaining usable life, to output the health score and predicted remaining usable life of each equipment component. Specifically:

[0074] A dynamic weight allocation strategy is adopted, and the weight coefficient is determined according to the ratio of the equipment's years of operation to its design life: the physical mechanism model has a higher weight in the early stage of operation, and the data-driven model has a higher weight in the later stage of operation due to the accumulation of actual degradation data. The health score and remaining usable life are respectively taken as the weighted average of the outputs of the two models.

[0075] When a model's output shows obvious anomalies (such as a sudden and significant jump in score exceeding the preset fluctuation range), the weight of that model is automatically reduced, and the result of another model is used as the primary reference. The final result is stored in the equipment health record and pushed to the maintenance decision module.

[0076] The historical operating data includes data from normal operating periods and data from the period before a fault occurred. Specifically, it refers to the time series of all operating parameters collected by IoT sensors at minute intervals since the equipment was put into operation. This data covers all operating conditions, including equipment startup, normal operation, partial load, and shutdown. The normal operating periods and the period before a known fault occurred are marked after preprocessing to train the model to identify fault symptoms.

[0077] Historical fault records include fault type and fault occurrence time. Specifically, they refer to a list of all fault repair records for the equipment extracted from the maintenance work order system or equipment ledger. Each record includes the fault occurrence time, fault location, fault type (e.g., bearing wear, insulation aging, blockage, leakage, etc.), fault level, and a brief repair description. These records are used to label training data and calculate actual remaining usable life.

[0078] The training steps for a data-driven model are as follows:

[0079] Data cleaning and processing of historical operational data, such as outlier removal and missing value imputation, are performed as follows:

[0080] Zero and null value segments caused by sensor power failure or communication interruption are removed. Outlier values ​​are identified and removed using statistical criteria. Short time periods with fewer than ten missing sampling points are filled using linear interpolation, while long time periods with missing data are marked and their features are ignored during training. The cleaned data is stored according to device number and time series.

[0081] The cleaned historical data is sliced ​​into fixed time windows, and multidimensional features are extracted within each time window. These features include: time-domain statistical features composed of mean, variance, peak factor, and kurtosis; frequency-domain features composed of peak frequencies and corresponding amplitudes extracted by fast Fourier transform; and time-series trend features composed of the slope of rising or falling trends extracted by linear fitting. Specifically:

[0082] For each parameter (such as current and temperature) within each time window, calculate the mean, variance, peak factor (maximum value divided by effective value), and kurtosis as time-domain statistical features. Perform a fast Fourier transform on the data within the window and extract the frequency and amplitude corresponding to the top five highest peaks in the amplitude spectrum as frequency-domain features.

[0083] Linear fitting is performed on the data points within the window, and the slope of the fitted line is taken as the trend feature (positive for upward trend, negative for downward trend). All features of all parameters are concatenated into a one-dimensional feature vector.

[0084] Using the feature vector as input, the known fault type within the corresponding time window is used as the classification label, and the difference between the timestamp of this time window and the fault occurrence time in historical fault records is used as the actual remaining usable life label, which is also used as the regression label. Specifically:

[0085] For each time window, generate two labels for the feature vector: classification label and category label. If the window timestamp falls within 24 hours before a certain fault occurs, label the type of the fault; otherwise, label it as normal.

[0086] Regression label: Subtract the window end timestamp from the fault occurrence time to obtain the actual remaining usable life (hours); the remaining usable life of normal category samples is set as the equipment design remaining life. Each sample consists of a feature vector, fault type label and remaining usable life label triplet, and the training set and validation set are divided proportionally.

[0087] The classification error is calculated using cross-entropy loss, and the regression error is calculated using mean squared error loss. The weighted sum of the two is used as the total loss function. The network parameters are iteratively optimized using the backpropagation algorithm until the total loss function converges, resulting in the trained data-driven model, which is as follows:

[0088] The network structure is a multi-task deep neural network, including a shared feature extraction layer (using multiple fully connected or one-dimensional convolutional layers), a classification branch (a classification output layer following a fully connected layer outputs the probability of each category), and a regression branch (a linear output layer following a fully connected layer outputs the remaining usable lifetime value).

[0089] The total loss function is composed of a weighted sum of classification error and regression error. The classification error uses cross-entropy loss to measure the difference between the predicted class and the true class, and the regression error uses mean squared error loss to measure the difference between the predicted lifetime and the true lifetime. A balance coefficient is set between the two to adjust the relative importance of the two losses.

[0090] Training employs an adaptive moment estimation optimizer, setting the initial learning rate and batch size, and performing multiple iterations. In each iteration, the classification accuracy and regression mean absolute error are evaluated on the validation set. Training is terminated early when the validation loss no longer decreases for several consecutive iterations. Finally, the model weight file with the best performance on the validation set is saved and called by the hybrid prediction evaluation module.

[0091] This implementation plan builds a mechanism model based on the equipment's own operating rules and conducts degradation simulations based on the actual operating conditions of the equipment. This allows for the determination of the equipment's status and lifespan based on its working logic. Furthermore, it trains a network model using the long-term operating information and fault records of the equipment to deeply explore potential operating patterns from multi-dimensional time-series information. The system relies on a multi-task network to simultaneously complete status identification and lifespan prediction. The two types of models dynamically adjust their weights according to the equipment's service progress to achieve result fusion. It can also automatically identify abnormal outputs and weaken the impact of interference. This fully combines the equipment's inherent operating logic with actual historical operating conditions, making the equipment status assessment and lifespan prediction more closely aligned with actual operating conditions and improving the reliability of operation and maintenance management.

[0092] Specifically, the preset maintenance triggering conditions employ a dual-judgment logic:

[0093] When the health score falls below a preset dynamic threshold, maintenance is triggered. This preset dynamic threshold is adaptively adjusted based on the equipment type, years of operation, and the criticality of the building space. Spaces with higher criticality have more conservative thresholds. Specifically:

[0094] The platform presets a baseline threshold for each type of equipment. The dynamic threshold is calculated as follows: the baseline threshold plus an adjustment value based on the number of years of operation (the adjustment value increases by a fixed number of points for each additional year of operation), and then multiplied by the criticality coefficient of the space in which it is located. Critical areas (such as data centers) have a higher space coefficient, general areas maintain the baseline coefficient, and non-critical areas have a lower coefficient. When the equipment health score falls below the calculated dynamic threshold, a maintenance warning is triggered.

[0095] When the predicted remaining usable life is less than the preset maintenance window duration, maintenance is deemed necessary. The preset maintenance window duration is the minimum lead time required for equipment maintenance preparation, determined based on spare parts procurement and personnel deployment cycles. Specifically:

[0096] The platform maintains the maintenance window duration parameter for each type of equipment. This parameter is the greater of the average spare parts procurement cycle and the lead time required for maintenance personnel deployment, plus a buffer time. The platform periodically checks the predicted remaining usable life of each piece of equipment. If it is less than the corresponding maintenance window duration, a maintenance work order is immediately triggered.

[0097] When any of the decision logics is satisfied, a predictive maintenance work order is generated, specifically as follows:

[0098] When any condition is met, a predictive maintenance work order is automatically created, which includes the trigger condition type, equipment component identification and name, spatial location, current health score and remaining usable life value. After the work order is created, its status is set to pending confirmation and it is pushed to the operation and maintenance management personnel for review. After the review is approved, it is transferred to the task pool and the execution order is determined by priority decision.

[0099] The specific steps for determining execution priority are as follows:

[0100] The health score and the remaining life expectancy prediction are standardized by using the maximum and minimum values ​​to map their corresponding values ​​between 0 and 1.

[0101] The standardized health scores and remaining useful life predictions are combined to generate maintenance priority scores for each equipment component. These scores are then sorted in descending order to generate a maintenance priority table, as follows:

[0102] The standardized health score and the remaining usable life prediction value are transformed using the reciprocal suppression mapping function f(x)=1 / (1+x), and then weighted according to a preset weight (e.g., both are 0.5) to obtain the maintenance priority score value of each equipment component. The scores are then arranged in descending order to generate a maintenance priority table.

[0103] Based on the maintenance priority table, the execution priority is determined as follows:

[0104] The work orders in the maintenance priority table are divided into three priority levels according to the range of the comprehensive maintenance priority score: a score of 0.8 or above is the emergency level, and maintenance personnel must be dispatched to the site immediately.

[0105] A score between 0.5 and 0.8 is considered high-level, and maintenance must be arranged within two hours.

[0106] A score below 0.5 is considered intermediate and can be processed on the same day.

[0107] If the number of work orders exceeds the available maintenance resources, emergency work orders will be assigned first, and the priority level will be written into the work order information.

[0108] This implementation plan, by setting up dual maintenance trigger logic and adaptively adjusting the judgment criteria based on the equipment's own condition and the scenario it is in, can meet the actual operation and maintenance needs of different equipment and different building areas, avoiding the problem of limited application scenarios for a unified judgment standard. At the same time, it sets maintenance windows based on spare parts procurement and personnel allocation to ensure that maintenance operations have sufficient preparation time. Finally, by comprehensively calculating maintenance priority scores through multi-dimensional indicators and classifying corresponding operation levels, it can reasonably distinguish the urgency of various maintenance tasks and accurately allocate operation and maintenance resources. This makes maintenance triggering more in line with actual working conditions, and operation and maintenance resource allocation more reasonable and efficient, thereby improving the level of refinement in building equipment operation and maintenance management.

[0109] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0110] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A building lifecycle intelligent maintenance management platform, characterized in that, include: The architectural design management unit is used to generate a design BIM model that carries design attributes and equipment information. The building construction management unit is used to generate a construction BIM model based on the design BIM model, and to receive construction data collected by IoT sensors during the construction phase, link it to the construction BIM model, and form an as-built BIM model. The intelligent operation and maintenance management unit includes an association mapping module, which is used to convert the as-built BIM model into an operation and maintenance BIM model and establish the data mapping relationship between each equipment component in the operation and maintenance BIM model and the corresponding IoT sensor. The hybrid prediction and evaluation module is used to acquire historical operating data, historical fault records and real-time operating status parameters of each equipment component, build a hybrid prediction model, and output the health score and remaining usable life prediction value of each equipment component. The maintenance decision module is used to generate predictive maintenance work orders and determine execution priorities when the health score or remaining usable life prediction result of any equipment component meets the preset maintenance trigger conditions.

2. The building lifecycle intelligent maintenance management platform according to claim 1, characterized in that, The specific steps for creating an as-built BIM model are as follows: Receive construction data collected by IoT sensors during the construction phase, and perform time-series alignment and outlier cleaning; Extract the unique identifier code of the construction BIM model, and establish an association mapping between the cleaned construction data and the construction BIM model according to the unique identifier code and the collection timestamp. Construction data is written into the construction BIM model in the form of an extended attribute set to generate the as-built BIM model.

3. The intelligent maintenance management platform for the entire building lifecycle as described in claim 1, characterized in that, The specific steps to convert the as-built BIM model into an operation and maintenance BIM model are as follows: Perform attribute cleaning on the as-built BIM model; Complete the operation and maintenance information of the as-built BIM model after attribute cleaning; The completed as-built BIM model is then lightweighted to obtain the operation and maintenance BIM model.

4. The building lifecycle intelligent maintenance management platform according to claim 3, characterized in that, The specific steps for establishing the data mapping relationship between each equipment component and its corresponding IoT sensor in the operation and maintenance BIM model are as follows: Extract the operation and maintenance asset codes and spatial coordinates of each equipment component in the operation and maintenance BIM model; Obtain the device identification and installation location information of each IoT sensor deployed within the building; Using the operation and maintenance asset code as the first matching key and the spatial coordinates as the second matching key, a one-to-one or one-to-many mapping relationship is established between each device component and the corresponding IoT sensor.

5. The intelligent maintenance management platform for the entire building lifecycle as described in claim 1, characterized in that, The hybrid prediction model is composed of a physical mechanism model of the equipment and a data-driven model. The specific steps for outputting the health score and remaining usable life prediction value of each equipment component are as follows: The physical mechanism model is based on the equipment operation mechanism. It uses real-time operating status parameters to simulate the degradation behavior of the equipment and outputs the first health score and the first remaining usable life. The data-driven model is trained using historical operating data and historical fault records. It takes real-time operating status parameters as input and outputs a second health score and a second remaining usable life. The hybrid prediction model performs weighted fusion of the first health score and the second health score, as well as the first remaining usable life and the second remaining usable life, to output the health score and remaining usable life prediction value of each equipment component.

6. The building lifecycle intelligent maintenance management platform according to claim 5, characterized in that, The training steps for a data-driven model are as follows: Perform data cleaning and processing on historical operational data; The historical operational data after data cleaning is sliced ​​by time window, and multidimensional features are extracted in each time window, including time domain statistical features, frequency domain features, and time series trend features. The feature vector is used as input, and the fault type corresponding to the time window is used as the classification label; The classification error is calculated using cross-entropy loss, and the regression error is calculated using mean squared error loss. The two are weighted and summed to obtain the total loss function. The network parameters are iteratively optimized through the backpropagation algorithm until the total loss function converges, thus obtaining the data-driven model.

7. The intelligent maintenance management platform for the entire building lifecycle as described in claim 1, characterized in that, The preset maintenance trigger conditions employ a dual-judgment logic: When the health score falls below a preset dynamic threshold, maintenance is required. When the predicted remaining usable life is less than the preset maintenance window duration, it is determined that maintenance needs to be triggered. When any of the decision logics is satisfied, a predictive maintenance work order is generated.

8. The intelligent maintenance management platform for the entire building lifecycle as described in claim 7, characterized in that, The specific steps for determining execution priority are as follows: The health score and remaining life expectancy prediction were standardized. The standardized health score and the remaining usable life prediction are combined to generate a maintenance priority score, which is then sorted in descending order to generate a maintenance priority table. Execution priorities are determined based on the maintenance priority table.

Images