AI-driven aviation consumable demand prediction and intelligent inventory optimization method and system

Abstract

The application discloses an AI-driven aviation consumable demand prediction and intelligent inventory optimization method and system, integrates multi-dimensional heterogeneous data such as flight dynamic scheduling, component service working conditions, 3D printing consumable attributes and a supply chain, constructs a collaborative prediction model through an AI algorithm, greatly improves the demand prediction accuracy compared with a traditional time series analysis method, effectively solves the problem of insufficient prediction accuracy in a complex aviation scene, and provides reliable data support for inventory optimization.

Description

Technical Field

[0001] This invention relates to the aerospace field, and more particularly to an AI-driven method and system for forecasting demand for aviation consumables and optimizing intelligent inventory. Background Technology

[0002] Aviation consumables specifically refer to 3D printing materials used in the aerospace field, primarily for manufacturing high-strength, high-temperature resistant, and corrosion-resistant high-performance components. Demand forecasting and intelligent inventory optimization provide scientific support for aviation consumables management by reducing costs, improving efficiency, optimizing resource allocation, and enhancing risk control, thus helping airlines achieve sustainable development.

[0003] Existing aviation consumables demand forecasting and intelligent inventory optimization systems mostly rely on traditional time series analysis or empirical formulas for demand forecasting. Inventory optimization strategies are mostly static threshold management, failing to fully integrate dynamic flight scheduling data and actual component service condition data, resulting in low forecast accuracy and difficulty in coping with complex and ever-changing aviation operation scenarios. At the same time, inventory optimization strategies are mostly static threshold management, lacking dynamic adjustment mechanisms based on real-time demand changes, supply chain response speed, and sudden failure maintenance needs. This can easily lead to inventory backlog of high-value consumables or temporary shortages of critical consumables, increasing warehousing costs and potentially affecting the timely repair and replacement of aviation components, thereby posing potential risks to flight safety and punctuality.

[0004] In conclusion, an AI-driven method and system for forecasting demand for aviation consumables and optimizing intelligent inventory is needed to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an AI-driven method and system for forecasting demand for aviation consumables and optimizing intelligent inventory, aiming to solve the aforementioned problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an AI-driven method for forecasting demand and optimizing intelligent inventory of aviation consumables, comprising the following steps:

[0007] Step S1: Multi-dimensional heterogeneous data collection, real-time collection of dynamic flight scheduling data, aviation component service status data, 3D printing consumable attribute data and supply chain data;

[0008] Step S2: Data preprocessing and feature engineering. The collected raw data is cleaned to remove noise values ​​and fill in missing items. Normalization is performed to unify the units and data format. Feature engineering is then performed to extract key features and generate a high-quality dataset that meets the input requirements of the AI ​​model.

[0009] Step S3: Multi-dimensional demand collaborative prediction. Based on the generated high-quality dataset, time series prediction model, multivariate regression model and deep learning prediction framework are used to extrapolate flight dynamic demand, component maintenance demand and 3D printing consumable demand respectively. The multi-task learning model is used to perform correlation analysis and bias elimination on the above multi-dimensional prediction results to generate a collaborative prediction scheme.

[0010] Step S4: Intelligent inventory dynamic optimization. Based on the generated collaborative prediction scheme, the intelligent inventory optimization algorithm is used to construct an optimization model with the core objectives of minimizing total inventory cost and maximizing inventory turnover rate, taking into account the life cycle cost of aerospace components, the storage loss characteristics of 3D printing consumables and supply chain response time constraints. The safety stock threshold, replenishment batch and pre-prepared stock quantity of various aerospace components and 3D printing consumables are dynamically calculated and adjusted.

[0011] Step S5: Feedback calibration and closed-loop iteration. Collect actual operational data and user feedback information after the implementation of the inventory optimization plan. Compare and analyze the actual data with the prediction and optimization results output by the system to identify the source of model deviation. Automatically trigger the dynamic calibration process of model parameters and algorithm optimization mechanism to continuously iterate the system performance.

[0012] Optionally, the collection of service condition data for aerospace components in step S1 specifically includes:

[0013] By using embedded high-precision MEMS sensors, fiber optic grating sensors, and piezoelectric vibration sensors, the temperature gradient distribution, vibration acceleration spectrum, hydraulic system pressure pulsation value, and structural stress deformation parameters of aerospace components are captured in real time during operation.

[0014] Optionally, the collection of 3D printing consumable attribute data in step S1 specifically includes:

[0015] It connects to the traceability database API of 3D printing consumables suppliers, laboratory material performance testing systems, and workshop storage environment monitoring terminals to obtain batch traceability information, melting temperature range, tensile strength, flexural modulus, heat distortion temperature, volume resistivity, water absorption rate, and temperature and humidity change curves of the storage environment for consumables.

[0016] Optionally, step S2 includes the following sub-steps:

[0017] Step S21: Multi-source heterogeneous data cleaning and denoising. The 3σ principle or isolated forest algorithm is used to identify and remove logical error records in flight dynamic data, instantaneous peak noise of sensors in component condition data, and outlier test values ​​in consumable attribute data. At the same time, linear interpolation, spline interpolation, or filling strategies based on historical mean are used to fill in missing data caused by transmission loss or sensor failure.

[0018] Step S22: Data normalization and standardization processing. For continuous numerical features of flight frequency and component vibration acceleration, the Min-Max normalization method is used to map them to the [0,1] interval; for features with specific distribution of physical and chemical properties of consumables, the Z-Score standardization method is used to transform them into a standard normal distribution; for route code and component model category data, one-hot encoding or tag encoding is performed to unify the data format.

[0019] Step S23: Time series data alignment and resampling. The high-frequency sampled component service condition data is resampled to a time granularity that matches the flight scheduling data and inventory records by using the sliding window averaging or maximum pooling method; a unified timestamp index is established to ensure that the flight status, component condition and consumable inventory data within the same time window are strictly aligned on the time axis.

[0020] Step S24: Key feature extraction and construction: Based on business logic and statistical patterns, extract and construct core feature vectors from the preprocessed data;

[0021] Step S25: Feature selection and dataset generation. Using the Pearson correlation coefficient matrix, mutual information method or tree-based feature importance assessment, select a subset of features that are strongly related to the demand prediction target, and remove multicollinear features and low-variance redundant features. Reorganize the selected feature vectors in time series order, divide them into training set, validation set and test set, and generate a structured dataset.

[0022] Optionally, the multi-dimensional demand collaborative prediction in step S3 includes the following sub-steps:

[0023] S31: Analyze flight schedule changes and delays using ARIMA time series models and LSTM deep learning models to predict flight frequency fluctuations and the emergency replacement needs of corresponding components;

[0024] S32: Using a multivariate regression model and a CNN-LSTM fusion framework, based on the temperature gradient and vibration acceleration spectrum characteristics in the component service condition data, the remaining service life, potential failure risk level and optimal maintenance window of the component are inferred.

[0025] S33: Combining the Transformer deep learning model with multivariate time series prediction algorithm, integrating consumable attribute data and consumption statistics, to predict the future demand, peak consumption periods and inventory warning thresholds for various types of consumables.

[0026] S34: Utilize the shared underlying features of the multi-task learning model to perform weighted fusion of the prediction results from S31 to S33, generating a collaborative prediction scheme covering production scheduling, maintenance arrangements, and inventory management.

[0027] Optionally, the intelligent inventory dynamic optimization in step S4 specifically includes the following sub-steps:

[0028] Step S41: Use genetic algorithm and particle swarm optimization algorithm to construct a component inventory optimization sub-model, and dynamically calculate the safety stock threshold, optimal replenishment batch and replenishment cycle of various aviation components;

[0029] Step S42: Using a multi-objective programming model, combined with the storage loss characteristics and remaining expiration date warning data of 3D printing consumables, determine the pre-production quantity of consumables, inventory layout adjustment plan, and priority use strategy for consumables nearing their expiration date;

[0030] Step S43: Based on the core objectives of minimizing total inventory cost and maximizing inventory turnover, a multi-objective optimization algorithm is used to solve the Pareto front, balance the emergency supply guarantee of components with the control of consumable storage costs, and generate a globally optimal inventory coordination adjustment plan.

[0031] Step S44: Connect with real-time inventory dynamic change data and supply chain status update information, and dynamically adjust the input parameters of the inventory optimization model.

[0032] Optionally, the actual operational data in step S5 includes actual inventory turnover rate, number of replenishment delays, and 3D printing consumable waste rate; user feedback information includes demand forecast deviation reports, evaluation of the effectiveness of inventory optimization schemes, and suggestions for function usage.

[0033] Optionally, the method may also include a visual interaction step:

[0034] It provides a visual interactive interface that allows users to query the demand forecast data generated in step S3, the dynamic results of inventory optimization generated in step S4, and the running status details of each module in real time, and to receive feedback information submitted by users through this interface.

[0035] An AI-driven aviation consumables demand forecasting and intelligent inventory optimization system, employing the AI-driven aviation consumables demand forecasting and intelligent inventory optimization method, includes a multi-dimensional heterogeneous data acquisition module, a data preprocessing and feature engineering module, a multi-dimensional demand collaborative forecasting module, an intelligent inventory dynamic optimization module, a feedback calibration and closed-loop iteration module, and a visualization interaction module.

[0036] The multi-dimensional heterogeneous data acquisition module is responsible for real-time acquisition of multi-source data, including dynamic flight scheduling data, service status data of aviation components, attribute data of 3D printing consumables, and supply chain data, and interfaces with various sensors, supplier APIs, laboratory testing systems, and workshop monitoring terminals.

[0037] The data preprocessing and feature engineering module is used to identify and remove logical errors, sensor noise and outliers, fill in missing data, unify the data format of different units, encode categorical data, resample data of different frequencies and align them to a unified time granularity, extract core features based on business logic, filter strongly correlated feature subsets, remove redundant features, and generate a high-quality structured dataset for use by AI models.

[0038] The multi-dimensional demand collaborative forecasting module is used to analyze flight plans, predict emergency replacement demand for components, extrapolate the remaining lifespan, failure risk and optimal maintenance window of components based on operating condition data, and predict future demand and peak periods by combining consumable attributes. The multi-task learning model is used to integrate the above multi-dimensional forecasting results, eliminate biases, and generate a collaborative forecasting solution covering production, maintenance and inventory.

[0039] The intelligent inventory dynamic optimization module is used to calculate the safety stock threshold, optimal replenishment batch and cycle for aerospace parts. It addresses the storage loss and shelf life of 3D printing consumables, aiming to minimize total cost and maximize turnover rate. It solves the Pareto front to generate a globally optimal inventory collaborative adjustment scheme, and dynamically updates and optimizes model parameters in real time by connecting with inventory and supply chain changes.

[0040] The feedback calibration and closed-loop iteration module is used to collect actual operational data and user feedback after the system is deployed, compare the actual data with the predicted optimization results, identify the source of deviation, automatically trigger the model parameter calibration and algorithm optimization mechanism, and continuously iterate and improve the system performance.

[0041] The visual interaction module provides a user interface that supports real-time querying of demand forecast data, inventory optimization results, and the running status of each module. It also receives user feedback and suggestions as input for system iteration.

[0042] The beneficial effects of this invention are:

[0043] 1. In this invention, multi-dimensional heterogeneous data such as flight dynamic scheduling, component service conditions, 3D printing consumables attributes and supply chain are integrated, and a collaborative prediction model is constructed through AI algorithms. Compared with traditional time series analysis methods, this greatly improves the accuracy of demand forecasting, effectively solves the problem of insufficient prediction accuracy in complex aviation scenarios, and provides reliable data support for inventory optimization.

[0044] 2. In this invention, based on real-time demand changes, supply chain response speed and emergency failure repair needs, the safety stock threshold, replenishment batch and pre-prepared stock quantity are dynamically adjusted through intelligent inventory optimization algorithm. Compared with the static threshold management mode, it can reduce the inventory backlog rate of high-value consumables, while reducing the occurrence rate of temporary shortage events of key consumables, thus balancing warehousing costs and emergency supply guarantee.

[0045] 3. In this invention, the actual operation data and user feedback are collected through the feedback module to achieve dynamic calibration of model parameters and algorithm optimization, ensuring that the system performance continues to improve with changes in business scenarios, thereby enhancing the system's adaptability and stability. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of a method flow of the present invention.

[0047] Figure 2 This is a schematic diagram of step S2 of the present invention.

[0048] Figure 3 This is a schematic diagram of step S3 of the present invention.

[0049] Figure 4 This is a schematic diagram of step S4 of the present invention.

[0050] Figure 5 This is a schematic diagram of a system structure according to the present invention. Detailed Implementation

[0051] To more clearly illustrate the technical solutions in the embodiments of the invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] like Figures 1 to 4 As shown, an AI-driven method for forecasting demand and optimizing intelligent inventory for aviation consumables includes the following:

[0053] Step S1: Multi-dimensional heterogeneous data collection, real-time collection of dynamic flight scheduling data, aviation component service status data, 3D printing consumable attribute data and supply chain data;

[0054] The system utilizes the dynamic scheduling data acquisition unit, component service condition data acquisition unit, 3D printing consumable attribute data acquisition unit, and consumable consumption statistics module of the data acquisition module to collect flight dynamic scheduling data, component service condition data, 3D printing consumable attribute data, and consumable consumption statistics in real time, respectively. Among them, the dynamic scheduling data acquisition unit uses OCR technology to identify unstructured information in paper scheduling logs, and the component service condition data acquisition unit captures multi-dimensional operating parameters through embedded high-precision MEMS sensors, fiber optic grating sensors, and piezoelectric vibration sensors.

[0055] Step S2: Data preprocessing and feature engineering. The collected raw data is cleaned to remove noise values ​​and fill in missing items. Normalization is performed to unify the units and data format. Feature engineering is then performed to extract key features and generate a high-quality dataset that meets the input requirements of the AI ​​model.

[0056] The data processing module receives the collected data and sequentially performs noise removal and missing item imputation through the data cleaning unit, the data normalization unit completes the standardization transformation of data of different dimensions, and the feature engineering unit extracts key features with high value for decision-making, generating a high-quality dataset that meets the input requirements of the AI ​​model.

[0057] Step S3: Multi-dimensional demand collaborative prediction. Based on the generated high-quality dataset, time series prediction model, multivariate regression model and deep learning prediction framework are used to extrapolate flight dynamic demand, component maintenance demand and 3D printing consumable demand respectively. The multi-task learning model is used to perform correlation analysis and bias elimination on the above multi-dimensional prediction results to generate a collaborative prediction scheme.

[0058] The demand forecasting module loads the high-quality dataset and uses the flight dynamic demand forecasting unit to deduce flight frequency fluctuations and emergency component replacement needs, the component maintenance demand forecasting unit to calculate the remaining service life and failure risk level of components, and the 3D printing consumables demand forecasting unit to predict the consumables demand and peak periods. Then, the multi-dimensional demand collaborative forecasting unit eliminates single-dimensional biases and outputs collaborative demand forecasting results covering production scheduling, maintenance arrangements, and inventory management.

[0059] Step S4: Intelligent inventory dynamic optimization. Based on the generated collaborative prediction scheme, the intelligent inventory optimization algorithm is used to construct an optimization model with the core objectives of minimizing total inventory cost and maximizing inventory turnover rate, taking into account the life cycle cost of aerospace components, the storage loss characteristics of 3D printing consumables and supply chain response time constraints. The safety stock threshold, replenishment batch and pre-prepared stock quantity of various aerospace components and 3D printing consumables are dynamically calculated and adjusted.

[0060] Based on the collaborative demand forecast results, the inventory optimization module calls the aviation component inventory optimization unit to calculate the component safety stock threshold and replenishment parameters, and the 3D printing consumables inventory optimization unit to determine the pre-production quantity of consumables. Through the multi-objective inventory collaborative optimization unit, it balances the emergency supply guarantee of components and the storage cost control of consumables, generates the globally optimal inventory adjustment plan, and the inventory parameter dynamic update unit connects with the latest data in real time to adjust the model input parameters.

[0061] Step S5: Feedback calibration and closed-loop iteration. Collect actual operational data and user feedback information after the implementation of the inventory optimization plan. Compare and analyze the actual data with the prediction and optimization results output by the system to identify the source of model deviation. Automatically trigger the dynamic calibration process of model parameters and algorithm optimization mechanism to continuously iterate the system performance.

[0062] The feedback module collects user feedback and actual operational data after the implementation of the inventory optimization plan through the information receiving unit. After being structured by the information classification unit, the deviation comparison and analysis unit compares the actual data with the system output results to locate the source of the deviation. If it is a parameter adaptation problem, the dynamic calibration unit automatically triggers the model parameter adjustment process. If it is a systematic deviation, the algorithm optimization trigger unit starts the algorithm optimization mechanism. Finally, the feedback closed-loop verification unit compares the core indicators before and after the adjustment (such as prediction accuracy and inventory cost reduction rate) to verify the optimization effect. If it fails, the deviation analysis and optimization process is repeated.

[0063] The method also includes the following steps:

[0064] Interactive Display: The interactive module presents the collaborative demand forecast results, inventory optimization schemes, and the operating status of each module through a visual dashboard. At the same time, the permission configuration unit assigns refined operation permissions to roles such as inventory managers and scheduling specialists. Users can obtain reports for specific scenarios through the data query and export unit. The operating status monitoring unit monitors the health of the modules in real time and triggers abnormal alarms. The early warning information push unit pushes early warning notifications such as inventory warnings, consumable expiration dates, and component failure risks to the corresponding responsible personnel.

[0065] System Integration and Security: The system interface module enables seamless data interaction with the airline's existing ERP system, flight scheduling system, and supply chain management system. The security module performs end-to-end encryption and access control during data transmission, storage, and access to ensure system data security.

[0066] like Figure 5 As shown, an AI-driven aviation consumables demand forecasting and intelligent inventory optimization system adopts the AI-driven aviation consumables demand forecasting and intelligent inventory optimization method, including a multi-dimensional heterogeneous data acquisition module, a data preprocessing and feature engineering module, a multi-dimensional demand collaborative forecasting module, an intelligent inventory dynamic optimization module, a feedback calibration and closed-loop iteration module, a visualization interaction module, and a system docking and security protection module.

[0067] The multi-dimensional heterogeneous data acquisition module is responsible for real-time acquisition of multi-source data, including dynamic flight scheduling data, service status data of aviation components, attribute data of 3D printing consumables, and supply chain data, and interfaces with various sensors, supplier APIs, laboratory testing systems, and workshop monitoring terminals.

[0068] The data preprocessing and feature engineering module is used to identify and remove logical errors, sensor noise and outliers, fill in missing data, unify the data format of different units, encode categorical data, resample data of different frequencies and align them to a unified time granularity, extract core features based on business logic, filter strongly correlated feature subsets, remove redundant features, and generate a high-quality structured dataset for use by AI models.

[0069] The multi-dimensional demand collaborative forecasting module is used to analyze flight plans, predict emergency replacement demand for components, extrapolate the remaining lifespan, failure risk and optimal maintenance window of components based on operating condition data, and predict future demand and peak periods by combining consumable attributes. The multi-task learning model is used to integrate the above multi-dimensional forecasting results, eliminate biases, and generate a collaborative forecasting solution covering production, maintenance and inventory.

[0070] The intelligent inventory dynamic optimization module is used to calculate the safety stock threshold, optimal replenishment batch and cycle for aerospace parts. It addresses the storage loss and shelf life of 3D printing consumables, aiming to minimize total cost and maximize turnover rate. It solves the Pareto front to generate a globally optimal inventory collaborative adjustment scheme, and dynamically updates and optimizes model parameters in real time by connecting with inventory and supply chain changes.

[0071] The feedback calibration and closed-loop iteration module is used to collect actual operational data and user feedback after the system is deployed, compare the actual data with the predicted optimization results, identify the source of deviation, automatically trigger the model parameter calibration and algorithm optimization mechanism, and continuously iterate and improve the system performance.

[0072] The visual interaction module provides a user interface that supports real-time querying of demand forecast data, inventory optimization results, and the running status of each module. It also receives user-submitted feedback and suggestions as input for system iteration.

[0073] The system integration and security protection module is used for seamless data interaction with the airline's existing ERP system, flight scheduling system and supply chain management system. The security module performs end-to-end encryption and access control for data transmission, storage and access processes to ensure system data security.

[0074] The data acquisition module collects real-time data on dynamic flight scheduling, service status of aviation components, attribute data of 3D printing consumables, and supply chain data according to a preset frequency or event triggering mechanism. After unifying and encapsulating the multi-source heterogeneous data, it transmits it to the data processing module.

[0075] The data processing module performs noise filtering and missing value imputation cleaning operations on the received raw data in sequence, completes the normalization processing of unit standardization and format unification, and extracts key features such as flight frequency, cumulative service time of components, and remaining validity period of consumables through feature engineering to generate a structured high-quality dataset, which is then pushed to the demand forecasting module.

[0076] The demand forecasting module loads high-quality datasets and automatically selects appropriate AI algorithms based on the demand scenario (such as using time series models for short-term demand, multivariate regression models for demand under multiple factors, and deep learning frameworks for complex nonlinear demand). It accurately extrapolates the future demand quantity, demand time window, and pre-stocking demand of 3D printing consumables for different types of aviation components, and outputs multi-dimensional demand forecasting results to the inventory optimization module.

[0077] The inventory optimization module takes demand forecast results as input, combines the life cycle cost of aerospace components, the storage loss characteristics of 3D printing consumables, and supply chain response time constraints, and calls genetic algorithms or particle swarm optimization algorithms to construct a multi-objective optimization model to solve for the optimal solution that minimizes total inventory cost and maximizes inventory turnover rate. It dynamically adjusts the safety stock threshold, replenishment batch, and pre-production quantity of 3D printing consumables for various components, generates inventory optimization solutions, and synchronizes them to the interaction module. At the same time, the interaction module displays demand forecast data and inventory optimization solutions in the form of visual charts (such as line charts, heat maps, and dashboards), allowing inventory managers and dispatchers to query details, export reports, or initiate manual intervention.

[0078] The feedback module collects real-time operational data after the implementation of the inventory optimization plan (such as actual inventory turnover rate, replenishment delays, and 3D printing consumable waste rate), and receives feedback information submitted by users through the interaction module. It performs deviation analysis between the system output results and the actual data. If the deviation exceeds the preset threshold, it triggers the model parameter calibration of the demand forecasting module or the algorithm constraint adjustment of the inventory optimization module to achieve continuous iterative optimization of the system.

[0079] The specific use case for this system is as follows:

[0080] Scenario 1: Daily inventory dynamic monitoring and precise optimization

[0081] Aviation company inventory managers can use the interactive module's visual dashboard to view real-time demand trend curves for core components such as engine high-pressure turbine blades and landing gear hydraulic components, output by the demand forecasting module, for the next 14 days. When the weekly demand forecast for a certain type of turbine blade increases by 23% compared to the previous week, the system's inventory optimization module automatically triggers a genetic algorithm. This algorithm combines the component's lifecycle cost (e.g., replacement labor costs account for 35%), the storage loss rate of 3D-printed titanium alloy consumables (recent batch loss rate is 2.8%), and supply chain response time (local supplier replenishment requires 24 hours) to quickly generate an optimization plan: increasing the safety stock threshold from 6 to 9 units, adjusting the replenishment batch from 10 to 12 units, and simultaneously increasing the pre-prepared 3D-printed consumables inventory by 18 kg. It also prioritizes using batches of consumables nearing their expiration date to reduce waste. The feedback module subsequently tracks the effectiveness of the plan. If the actual inventory turnover rate increases by 12% and the consumables waste rate decreases by 3%, the cost weight parameters in the model are automatically recalibrated.

[0082] Scenario 2: Emergency Response to Sudden Flight Scheduling Adjustments

[0083] After receiving notification of five additional extra flights on popular domestic routes, the flight dispatcher immediately captures the flight schedule change data in the system's data acquisition module and synchronizes it to the data processing module for feature extraction (such as the route type and aircraft type matching component requirements of the extra flights). The demand forecasting module updates the component demand forecast for the next 7 days within 10 minutes (e.g., an increase of 15 hinges for a certain model of cabin door). The inventory optimization module uses a particle swarm optimization algorithm, considering the air freight cost of emergency replenishment (4 times higher than regular land transport) and the real-time production capacity of 3D printing, recommending prioritizing the use of pre-prepared goods at the local 3D printing center (which can complete printing within 4 hours) to reduce the frequency of emergency air freight. The interaction module pushes the optimization plan to the dispatcher and 3D printing workshop administrator in real time. The feedback module subsequently collects the actual waste rate of printing consumables (if it is 1.2%, lower than the standard value of 1.5%) and updates the loss characteristic parameters of that batch of consumables into the model.

[0084] Scenario 3: Intelligent Management and Loss Optimization of 3D Printing Consumables

[0085] When a 3D printing workshop administrator queries the inventory status of a batch of aluminum alloy powder via the interactive module, the system's data processing module has already integrated the loss data from the last 10 printing jobs (the average loss rate for this batch is 3.2%, higher than the historical average of 2.5%). The demand forecasting module, considering the printing workload for the next two weeks (requiring 200kg of this powder), and the inventory optimization module, using a multi-objective programming model to balance storage costs and loss risks, recommend reducing the pre-prepared quantity of this powder by 12% and adjusting the humidity threshold of the storage environment (from 40% to 35%). The feedback module subsequently collects the actual waste rate data for this powder weekly. If it decreases to 2.6% for three consecutive weeks, it automatically optimizes the storage loss characteristic function in the model, improving the accuracy of subsequent predictions.

[0086] Scenario 4: Rapid Iteration of Inventory Strategies Under Supply Chain Anomalies

[0087] The supply chain department discovered that a port closure caused by a typhoon affected an overseas supplier, extending the replenishment response time from 72 hours to 144 hours. The system's data acquisition module updated supply chain constraint parameters in real time, the demand forecasting module extended the long-term demand forecast window from 30 days to 60 days, and the inventory optimization module employed a multi-objective programming model to find the optimal solution between "minimizing inventory costs" and "avoiding stockout risk": increasing the safety stock threshold for landing gear seals supplied by this supplier by 15%, while simultaneously increasing the stock of locally 3D-printed replacement seals (accounting for 20% of total demand). The interaction module immediately pushed an adjustment notification to inventory management personnel, and the feedback module subsequently recorded the number of stockouts after implementing this solution (reducing from the expected 2 times to 0 times), incorporating this abnormal scenario's response strategy into the model's case library to improve the system's response efficiency to similar events.

[0088] These scenarios cover core aspects of daily operations, emergency adjustments, consumables management, and supply chain anomalies for aviation companies. Through the collaborative operation of various modules in the system, a closed-loop intelligent management system is achieved, from data collection to model optimization, effectively solving pain points in aviation consumables inventory management.

[0089] Specifically, the data acquisition module includes:

[0090] The dynamic scheduling data acquisition unit connects to the civil aviation data exchange platform and the airport scheduling system API to acquire structured data such as flight plan changes, delay duration classification records, and temporary route rerouting instructions in real time, and uses OCR technology to identify unstructured information in paper scheduling logs.

[0091] The component service condition data acquisition unit, through embedded high-precision MEMS sensors, fiber optic grating sensors and piezoelectric vibration sensors, captures multi-dimensional service condition parameters such as temperature gradient distribution, vibration acceleration spectrum, hydraulic system pressure pulsation value, and structural stress deformation in real time during operation.

[0092] The 3D printing consumables attribute data acquisition unit connects to the traceability database API of 3D printing consumables suppliers, the laboratory material performance testing system, and the workshop storage environment monitoring terminal to obtain in real time batch traceability information of consumables, physical and chemical property parameters tested in the laboratory (such as melting temperature range, tensile strength, flexural modulus, heat distortion temperature, volume resistivity, water absorption), temperature and humidity change curves of the storage environment, and expiration date warning data.

[0093] The consumables consumption statistics module is used to collect real-time statistics on the actual consumption, waste rate, and dynamic changes in inventory of consumables for each 3D printing production task, providing data support for consumables inventory management and procurement decisions.

[0094] With the coordinated efforts of the aforementioned modules, the data acquisition module can achieve comprehensive coverage, real-time capture, and efficient integration of multi-source heterogeneous data, such as flight scheduling, component service conditions, 3D printing consumable attributes, and consumption statistics. This ensures that the collected data covers the key dimensions required for the entire process of aviation consumable demand forecasting and intelligent inventory optimization, and possesses high timeliness, completeness, and accuracy. This provides solid and reliable raw data support for the subsequent data processing modules' cleaning, normalization, and feature engineering operations, facilitating accurate demand projection and inventory strategy optimization for subsequent modules of the system.

[0095] Specifically, the data acquisition and processing module includes:

[0096] The data cleaning unit is used to perform fine-grained processing on the collected data, such as removing noise values ​​and filling in missing items.

[0097] Data normalization unit performs standardization operations on data types of different dimensions;

[0098] The feature engineering unit is used to extract, construct, and screen key features that are of high value for intelligent scheduling and component maintenance decisions of aerospace 3D printing consumables from cleaned and normalized multi-source data, providing a core input feature set for subsequent AI model training.

[0099] With the coordinated efforts of the aforementioned modules, the data acquisition and processing module can transform multi-source heterogeneous raw data (including flight dynamic scheduling, component service conditions, 3D printing consumable attributes and consumption statistics) into high-quality input datasets with unified format, accurate feature dimensions, and strong business relevance. This effectively eliminates data noise, fills information gaps, and extracts core decision features, ensuring that the output dataset fully adapts to the input specifications of the time series prediction model, multivariate regression model, and deep learning prediction framework in the demand forecasting module. This provides reliable data support for subsequent accurate multi-dimensional demand extrapolation and intelligent inventory optimization, greatly improving the overall prediction accuracy and optimization efficiency of the system.

[0100] Specifically, the demand forecasting module includes:

[0101] The flight dynamic demand forecasting unit uses the ARIMA time series model and the LSTM deep learning model to analyze the cleaned and normalized flight schedule changes and delay duration classification records, predict flight frequency fluctuations, route adjustment trends and corresponding emergency replacement needs of parts, and provide a basis for priority scheduling of 3D printing production tasks for aviation parts.

[0102] The component maintenance requirement prediction unit employs a multivariate regression model and a CNN-LSTM fusion framework. The CNN-LSTM framework processes vibration spectrum data by first converting the collected vibration acceleration spectrum data into a two-dimensional spectrogram (time-frequency dimension), followed by frame segmentation and standardization preprocessing. Then, multi-scale convolutional kernels (such as 1D or 2D kernels) of the CNN are used to perform layer-by-layer convolution operations on the spectrogram, extracting local spatial features (including key information such as the frequency location of abnormal vibration peaks, mode distribution, and energy changes). Pooling layers are used to reduce the dimensionality and select features, retaining the most discriminative spatial feature sequences. These feature sequences are then input into an LSTM layer, where the LSTM's gating mechanism captures the dynamic evolution of the feature sequences over time, such as the cumulative trend of abnormal vibration modes and the time dependence of component performance degradation. Finally, a fully connected layer and a softmax activation function output the probability range of component failure or the predicted remaining service life, thereby accurately determining whether components need early maintenance or replacement, providing core data support for the quantitative extrapolation of component maintenance requirements.

[0103] The input tensor dimension of the CNN-LSTM fusion framework is [batch_size, sequence_len, freq_dim, time_dim, 1], where batch_size is the sample batch size during model training or prediction, sequence_len represents the length of the time series composed of continuous vibration spectrum frames (i.e., the number of frames contained within the covered monitoring time window), freq_dim corresponds to the number of frequency dimension points in the two-dimensional spectrogram, time_dim corresponds to the number of time dimension points in the spectrogram, and the last dimension 1 indicates that the spectrogram is single-channel grayscale image data. This dimension design can accurately adapt to the spatial feature extraction and temporal dynamic pattern capture requirements of the CNN-LSTM fusion framework for vibration spectrum data, ensuring that the model efficiently processes local spatial anomaly patterns and long-term temporal dependencies in the spectrum, providing structurally reasonable input data support for accurate prediction of component remaining service life and failure probability; based on the key features of temperature gradient and vibration acceleration spectrum in component service condition data, the remaining service life, potential failure risk level, and optimal maintenance window period of the component can be deduced, helping to deploy preventive maintenance measures in advance;

[0104] The 3D printing consumables demand forecasting unit combines the Transformer deep learning model and multivariate time series forecasting algorithm to integrate consumables attribute data, consumption statistics data and inventory dynamic information to predict the future demand for various types of consumables, peak consumption periods and inventory warning thresholds, providing data support for consumables procurement planning and inventory optimization.

[0105] The multi-dimensional demand collaborative forecasting unit uses a multi-task learning model to perform correlation analysis on the forecast results of three types: flight dynamics, component maintenance, and consumable demand. This eliminates the bias of single-dimensional forecasts and generates a collaborative forecasting scheme covering production scheduling, maintenance arrangements, and inventory management, achieving global optimal matching of multi-source demand in the aviation field.

[0106] Specifically, the multi-task learning model employs a Transformer-based multi-task learning framework with a shared encoder and multi-task decoder. This framework performs deep feature extraction on the high-quality dataset output by the data processing module through a low-level shared encoder, generating general feature representations covering multiple dimensions such as flight dynamics, component maintenance, and consumable demand. For each of the three sub-tasks, task-specific decoder branches are designed: the flight dynamics branch introduces a time attention layer to accurately capture the temporal dependency between flight frequency and route adjustments; the component maintenance branch incorporates a condition feature attention module to strengthen the weights of key parameters such as temperature gradient and vibration acceleration spectrum; and the consumable demand branch combines an inventory constraint attention mechanism to optimize the prediction accuracy of peak consumable consumption periods and warning thresholds. Finally, end-to-end training is achieved through a joint loss function (weighted fusion of temporal prediction loss, regression loss, and classification loss), ensuring the prediction accuracy of each sub-task while strengthening feature sharing and collaborative relationships between tasks.

[0107] The weighting method of the loss function is as follows:

[0108] ,

[0109] In the formula, Total loss;

[0110] For time series prediction loss, for Weighting coefficients;

[0111] To regress the loss, for Weighting coefficients;

[0112] For classifying losses, for Weighting coefficients;

[0113] And satisfy .

[0114] The initial weights are assigned based on aviation-related business priorities: component maintenance tasks are directly related to flight safety, hence the classification loss... Initial weights for (predicted component failure risk levels) Set to 0.4; Flight dynamic demand forecasting impacts production scheduling efficiency, resulting in time-series losses. (Predicted fluctuations in corresponding flight frequencies) Initial weights Set to 0.3; 3D printing consumables demand forecast relationship with inventory cost, regression loss (Corresponding to the predicted demand for consumables) Initial weights Set to 0.3. Compared with traditional single-task prediction models, this model not only reduces parameter redundancy and improves computational efficiency, but also effectively eliminates the bias of single-dimensional prediction, generating a global collaborative prediction solution that is more in line with the actual business scenarios in the aviation field, providing accurate and comprehensive multi-dimensional demand input for the inventory optimization module.

[0115] Through the coordinated efforts of the aforementioned modules, the demand forecasting module can accurately output multi-dimensional collaborative forecasting results covering emergency flight scheduling needs, component maintenance plan needs, and 3D printing consumable usage needs. This effectively improves the accuracy of single-dimensional forecasts and the correlation between multi-dimensional needs. It provides a precise demand input foundation for the inventory optimization module to build an optimization model with the core objectives of minimizing total inventory costs and maximizing inventory turnover. At the same time, it provides scientific data support for core business processes such as production scheduling, maintenance arrangements, and inventory management for aviation enterprises. This enables enterprises to respond in advance to demand fluctuations and potential risks, ensuring the stable and efficient operation of the aviation supply chain.

[0116] Specifically, the inventory optimization module includes:

[0117] The aviation component inventory optimization unit integrates the life cycle cost data of aviation components, supply chain response time parameters, and emergency replacement demand of components output by the demand forecasting module. It uses genetic algorithm and particle swarm optimization algorithm to construct a component inventory optimization sub-model, dynamically calculates the safety stock threshold, optimal replenishment batch and replenishment cycle of various aviation components, and provides accurate decision support for the inventory turnover and emergency supply of aviation components.

[0118] The 3D printing consumables inventory optimization unit combines the storage loss characteristics of 3D printing consumables, the remaining expiration date warning data, and the demand and peak consumption periods output by the consumables demand forecasting unit. It uses a multi-objective programming model to determine the pre-prepared stock quantity of consumables, the inventory layout adjustment plan, and the priority use strategy for consumables with near expiration dates, effectively reducing consumables storage waste and expiration risk.

[0119] The multi-objective inventory collaborative optimization unit, based on the core objectives of minimizing total inventory cost and maximizing inventory turnover rate, integrates the inventory optimization sub-model results of aviation parts and consumables, and uses a multi-objective optimization algorithm to balance emergency supply guarantee of parts and storage cost control of consumables, generating a globally optimal inventory collaborative adjustment scheme covering aviation parts and 3D printing consumables.

[0120] The inventory parameter dynamic update unit connects in real time with the data acquisition and processing module to obtain dynamic inventory change data, the latest forecast results from the demand forecasting module, and supply chain status updates. It dynamically adjusts the input parameters of the inventory optimization model (such as demand fluctuation coefficient and supply chain response time deviation) to ensure that the inventory optimization solution can quickly adapt to changes in actual business scenarios and maintain the efficiency and flexibility of inventory management.

[0121] With the coordinated efforts of the above modules, the inventory optimization module can accurately output a global collaborative optimization solution that covers the safety stock threshold for aerospace components, the optimal replenishment batch and the pre-prepared stock quantity of 3D printing consumables, and the priority use strategy for consumables nearing their expiration date. This effectively balances the core needs of emergency supply guarantee for aerospace components and cost control of 3D printing consumables storage, achieving a significant reduction in total inventory costs and a steady increase in inventory turnover.

[0122] Specifically, the interaction module includes:

[0123] The visualization dashboard integrates the multi-dimensional forecast results of the demand forecasting module, the dynamic adjustment scheme of the inventory optimization module, and the key feature data of the data acquisition and processing module. It uses visualization components such as line charts, bar charts, heat maps, and dashboards to intuitively present flight dynamic trends, component maintenance demand distribution, consumable inventory warning status, and global inventory cost optimization effects. It also supports user-defined chart combinations and data display dimensions.

[0124] The permission configuration unit sets up refined permission control rules for different roles such as inventory management personnel and scheduling specialists. For example, inventory management personnel can access the consumables inventory adjustment interface and procurement decision data, while scheduling specialists can view flight dynamic forecast results and production task priority scheduling information, ensuring the security and compliance of data access and operation.

[0125] The data query and export unit provides multi-dimensional filtering conditions (such as time range, consumable type, component number, flight route, etc.) to enable users to accurately query raw collected data, processed datasets, demand forecast reports and inventory optimization solutions in specific scenarios. It also supports exporting query results to Excel, PDF, CSV and other formats to meet the needs of business report generation and data analysis.

[0126] The operation status monitoring unit collects the operation parameters (such as data processing throughput, model prediction accuracy, and task execution time) of each module (data acquisition module, data acquisition and processing module, demand forecasting module, and inventory optimization module) in real time. It intuitively displays the health status of the module through status indicator lights and abnormal alarm prompts. When abnormalities such as data acquisition interruption or excessive model prediction deviation occur, it automatically triggers audible and visual alarms or pop-up alarms.

[0127] The operation log recording unit records in detail all users' login information, operation behaviors (such as data query, parameter adjustment, and warning confirmation), operation time, and operation results, forming a traceable operation audit log, which facilitates system maintenance personnel to troubleshoot problems and business management personnel to conduct compliance checks.

[0128] The early warning information push unit connects with the inventory warning threshold, consumable expiration warning, and component failure risk level data of the demand forecasting module. It pushes early warning notifications to the relevant responsible personnel in real time through SMS, corporate email, system built-in pop-ups, or mobile APP push, such as consumable inventory being lower than the safety threshold, a batch of consumables being about to expire, or components having insufficient remaining service life, to help with rapid response and decision-making.

[0129] With the synergy of the above modules, the interaction module can provide different roles such as inventory managers and dispatchers in aviation enterprises with intuitive and efficient visual decision support, refined access control, convenient data query and export services, real-time system operation monitoring and anomaly alarms, traceable operation auditing, and accurate early warning information push, realizing a user-friendly interactive experience throughout the entire process from data presentation to business operation.

[0130] Specifically, the feedback module includes:

[0131] The information receiving unit is used to receive user feedback information such as demand forecast deviation reports, inventory optimization plan implementation effect evaluations, and function usage suggestions submitted through the interactive module.

[0132] The information classification unit automatically categorizes and tags the feedback according to the feedback type (model performance, functional experience, business adaptation), providing a structured data foundation for subsequent deviation analysis and optimization.

[0133] The actual operation data collection unit collects real-time operational indicator data after the implementation of the inventory optimization plan, including actual inventory turnover rate, number of consumable replenishment delays, actual waste rate of 3D printing consumables, emergency replacement response time of parts, and accuracy of procurement decision execution, forming an actual data benchmark library corresponding to the system's prediction and optimization results.

[0134] The deviation comparison and analysis unit performs multi-dimensional quantitative comparisons of user feedback deviation information, actual operational data, and system output demand forecast results and inventory optimization solutions. It calculates the deviation rate and locates the source of deviation (such as missing key features in the feature engineering process, insufficient adaptability of prediction model parameters, and disconnect between inventory optimization constraints and actual business scenarios).

[0135] The dynamic calibration unit automatically triggers the parameter adjustment process of the AI ​​model based on the deviation tracing results. For example, it adjusts the time step of the LSTM model in the demand forecasting module, the number of attention heads of the Transformer model, or the crossover and mutation probability of the genetic algorithm in the inventory optimization module, so as to achieve adaptive optimization of model parameters.

[0136] The algorithm optimization trigger unit automatically initiates the algorithm optimization mechanism for systematic deviations that cannot be resolved through parameter calibration (such as changes in the algorithm framework to adapt to different scenarios). This includes updating the feature selection logic of feature engineering, replacing the objective function weights in the inventory optimization model, and introducing a reinforcement learning framework to optimize multi-objective inventory collaboration strategies.

[0137] The feedback closed-loop verification unit verifies the effectiveness of the calibrated model and the optimized algorithm. By comparing key indicators such as prediction accuracy, inventory cost reduction rate, and inventory turnover rate improvement before and after the adjustment, it confirms whether the deviation has been effectively corrected. If the verification fails, the deviation analysis and optimization process is restarted.

[0138] With the synergistic cooperation of the above modules, the feedback module can achieve deep linkage between user feedback information and actual operational data, and build a complete closed-loop mechanism from problem discovery and deviation tracing to model calibration and algorithm iteration.

[0139] Specifically, it also includes a system interface module, used to achieve seamless integration with the airline's existing ERP system, flight scheduling system, and supply chain management system. This means that the system interface module enables bidirectional real-time synchronization of core business data such as purchase orders and inventory ledgers from the airline's existing ERP system, real-time flight schedule changes and delay information from the flight scheduling system, and logistics and transportation status and supplier response time from the supply chain management system.

[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI-driven method for forecasting demand and optimizing intelligent inventory for aviation consumables, characterized in that, Includes the following steps: Step S1: Multi-dimensional heterogeneous data collection, real-time collection of dynamic flight scheduling data, aviation component service status data, 3D printing consumable attribute data and supply chain data; Step S2: Data preprocessing and feature engineering. The collected raw data is cleaned to remove noise values ​​and fill in missing items. Normalization is performed to unify the units and data format. Feature engineering is then performed to extract key features and generate a high-quality dataset that meets the input requirements of the AI ​​model. Step S3: Multi-dimensional demand collaborative prediction. Based on the generated high-quality dataset, time series prediction model, multivariate regression model and deep learning prediction framework are used to extrapolate flight dynamic demand, component maintenance demand and 3D printing consumable demand respectively. The multi-task learning model is used to perform correlation analysis and bias elimination on the above multi-dimensional prediction results to generate a collaborative prediction scheme. Step S4: Intelligent inventory dynamic optimization. Based on the generated collaborative prediction scheme, the intelligent inventory optimization algorithm is used to construct an optimization model with the core objectives of minimizing total inventory cost and maximizing inventory turnover rate, taking into account the life cycle cost of aerospace components, the storage loss characteristics of 3D printing consumables and supply chain response time constraints. The safety stock threshold, replenishment batch and pre-prepared stock quantity of various aerospace components and 3D printing consumables are dynamically calculated and adjusted. Step S5: Feedback calibration and closed-loop iteration. Collect actual operational data and user feedback information after the implementation of the inventory optimization plan. Compare and analyze the actual data with the prediction and optimization results output by the system to identify the source of model deviation. Automatically trigger the dynamic calibration process of model parameters and algorithm optimization mechanism to continuously iterate the system performance.

2. The AI-driven method for forecasting demand and optimizing intelligent inventory of aviation consumables according to claim 1, characterized in that, The collection of service condition data for aerospace components in step S1 specifically includes: By using embedded high-precision MEMS sensors, fiber optic grating sensors, and piezoelectric vibration sensors, the temperature gradient distribution, vibration acceleration spectrum, hydraulic system pressure pulsation value, and structural stress deformation parameters of aerospace components are captured in real time during operation.

3. The AI-driven method for forecasting demand and optimizing intelligent inventory of aviation consumables according to claim 2, characterized in that, The collection of 3D printing consumable attribute data in step S1 specifically includes: It connects to the traceability database API of 3D printing consumables suppliers, laboratory material performance testing systems, and workshop storage environment monitoring terminals to obtain batch traceability information, melting temperature range, tensile strength, flexural modulus, heat distortion temperature, volume resistivity, water absorption rate, and temperature and humidity change curves of the storage environment for consumables.

4. The AI-driven method for forecasting demand and optimizing intelligent inventory of aviation consumables according to claim 1, characterized in that, Step S2 includes the following sub-steps: Step S21: Multi-source heterogeneous data cleaning and denoising. The 3σ principle or isolated forest algorithm is used to identify and remove logical error records in flight dynamic data, instantaneous peak noise of sensors in component condition data, and outlier test values ​​in consumable attribute data. At the same time, linear interpolation, spline interpolation, or filling strategies based on historical mean are used to fill in missing data caused by transmission loss or sensor failure. Step S22: Data normalization and standardization processing. For continuous numerical features of flight frequency and component vibration acceleration, the Min-Max normalization method is used to map them to the [0,1] interval; for features with specific distribution of physical and chemical properties of consumables, the Z-Score standardization method is used to transform them into a standard normal distribution; for route code and component model category data, one-hot encoding or tag encoding is performed to unify the data format. Step S23: Time series data alignment and resampling. The high-frequency sampled component service condition data is resampled to a time granularity that matches the flight scheduling data and inventory records by using the sliding window averaging or maximum pooling method; a unified timestamp index is established to ensure that the flight status, component condition and consumable inventory data within the same time window are strictly aligned on the time axis. Step S24: Key feature extraction and construction: Based on business logic and statistical patterns, extract and construct core feature vectors from the preprocessed data; Step S25: Feature selection and dataset generation. Using the Pearson correlation coefficient matrix, mutual information method or tree-based feature importance assessment, a subset of features strongly correlated with the demand prediction target is selected, and multicollinear features and low-variance redundant features are removed. The selected feature vectors are reorganized in time series order, and divided into training, validation and test sets to generate a structured dataset.

5. The AI-driven method for forecasting demand and optimizing intelligent inventory of aviation consumables according to claim 1, characterized in that, The multi-dimensional demand collaborative prediction in step S3 includes the following sub-steps: S31: Analyze flight schedule changes and delays using ARIMA time series models and LSTM deep learning models to predict flight frequency fluctuations and the emergency replacement needs of corresponding components; S32: Using a multivariate regression model and a CNN-LSTM fusion framework, based on the temperature gradient and vibration acceleration spectrum characteristics in the component service condition data, the remaining service life, potential failure risk level and optimal maintenance window of the component are inferred. S33: Combining the Transformer deep learning model with multivariate time series prediction algorithm, integrating consumable attribute data and consumption statistics, to predict the future demand, peak consumption periods and inventory warning thresholds for various types of consumables. S34: Utilize the shared underlying features of the multi-task learning model to perform weighted fusion of the prediction results from S31 to S33, generating a collaborative prediction scheme covering production scheduling, maintenance arrangements, and inventory management.

6. The AI-driven method for forecasting demand and optimizing intelligent inventory of aviation consumables according to claim 1, characterized in that, The intelligent inventory dynamic optimization in step S4 specifically includes the following sub-steps: Step S41: Use genetic algorithm and particle swarm optimization algorithm to construct a component inventory optimization sub-model, and dynamically calculate the safety stock threshold, optimal replenishment batch and replenishment cycle of various aviation components; Step S42: Using a multi-objective programming model, combined with the storage loss characteristics and remaining expiration date warning data of 3D printing consumables, determine the pre-production quantity of consumables, inventory layout adjustment plan, and priority use strategy for consumables nearing their expiration date; Step S43: Based on the core objectives of minimizing total inventory cost and maximizing inventory turnover, a multi-objective optimization algorithm is used to solve the Pareto front, balance the emergency supply guarantee of components with the control of consumable storage costs, and generate a globally optimal inventory coordination adjustment plan. Step S44: Connect with real-time inventory dynamic change data and supply chain status update information, and dynamically adjust the input parameters of the inventory optimization model.

7. The AI-driven method for forecasting demand and optimizing intelligent inventory of aviation consumables according to claim 1, characterized in that, The actual operational data in step S5 includes actual inventory turnover rate, number of replenishment delays, and 3D printing consumable waste rate; user feedback information includes demand forecast deviation reports, evaluation of the effectiveness of inventory optimization solutions, and suggestions for function usage.

8. The AI-driven method for forecasting demand and optimizing intelligent inventory of aviation consumables according to claim 1, characterized in that, The method also includes a visual interaction step: It provides a visual interactive interface that allows users to query the demand forecast data generated in step S3, the dynamic results of inventory optimization generated in step S4, and the running status details of each module in real time, and to receive feedback information submitted by users through this interface.

9. The AI-driven method for forecasting demand and optimizing intelligent inventory of aviation consumables according to claim 1, characterized in that, The method also includes system integration and security protection steps: Seamless data interaction with existing airline ERP systems, flight scheduling systems, and supply chain management systems; the security module performs end-to-end encryption and access control during data transmission, storage, and access to ensure system data security.

10. An AI-driven aviation consumables demand forecasting and intelligent inventory optimization system, employing the AI-driven aviation consumables demand forecasting and intelligent inventory optimization method as described in any one of claims 1-9, characterized in that, It includes a multi-dimensional heterogeneous data acquisition module, a data preprocessing and feature engineering module, a multi-dimensional demand collaborative forecasting module, an intelligent inventory dynamic optimization module, a feedback calibration and closed-loop iteration module, a visualization interaction module, and a system docking and security protection module. The multi-dimensional heterogeneous data acquisition module is responsible for real-time acquisition of multi-source data, including dynamic flight scheduling data, service status data of aviation components, attribute data of 3D printing consumables, and supply chain data, and interfaces with various sensors, supplier APIs, laboratory testing systems, and workshop monitoring terminals. The data preprocessing and feature engineering module is used to identify and remove logical errors, sensor noise and outliers, fill in missing data, unify the data format of different units, encode categorical data, resample data of different frequencies and align them to a unified time granularity, extract core features based on business logic, filter strongly correlated feature subsets, remove redundant features, and generate a high-quality structured dataset for use by AI models. The multi-dimensional demand collaborative forecasting module is used to analyze flight plans, predict emergency replacement demand for components, extrapolate the remaining lifespan, failure risk and optimal maintenance window of components based on operating condition data, and predict future demand and peak periods by combining consumable attributes. The multi-task learning model is used to integrate the above multi-dimensional forecasting results, eliminate biases, and generate a collaborative forecasting solution covering production, maintenance and inventory. The intelligent inventory dynamic optimization module is used to calculate the safety stock threshold, optimal replenishment batch and cycle for aerospace parts. It addresses the storage loss and shelf life of 3D printing consumables, aiming to minimize total cost and maximize turnover rate. It solves the Pareto front to generate a globally optimal inventory collaborative adjustment scheme, and dynamically updates and optimizes model parameters in real time by connecting with inventory and supply chain changes. The feedback calibration and closed-loop iteration module is used to collect actual operational data and user feedback after the system is deployed, compare the actual data with the predicted optimization results, identify the source of deviation, automatically trigger the model parameter calibration and algorithm optimization mechanism, and continuously iterate and improve the system performance. The visual interaction module provides a user interface that supports real-time querying of demand forecast data, inventory optimization results, and the running status of each module. It also receives user-submitted feedback and suggestions as input for system iteration. The system integration and security protection module is used for seamless data interaction with the airline's existing ERP system, flight scheduling system and supply chain management system. The security module performs end-to-end encryption and access control for data transmission, storage and access processes to ensure system data security.

Images