A key standard part complete set gap prediction method based on assembly pull
By constructing a combined LSTM and XGBoost model, the problem of unpredictable fluctuations in demand for key standard parts in final assembly-driven production was solved, enabling accurate identification and early warning of future shortages, and improving the accuracy and stability of material supply.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING UNIV OF SCI & TECH
- Filing Date
- 2026-04-27
- Publication Date
- 2026-07-21
AI Technical Summary
In the context of final assembly-driven production, existing technologies struggle to accurately predict future demand fluctuations for key standard components, making it difficult to identify gap risks and lacking early warning capabilities, which in turn affects the continuous progress of assembly nodes and material supply.
By constructing a combined model based on LSTM and XGBoost, and integrating order tasks, assembly node plans, manufacturing bill of materials, and key standard parts status data for unified processing, a key standard parts gap prediction model is built to achieve early identification and warning of future gaps.
It improves the accuracy and stability of critical standard component shortage prediction, enables early risk identification, supports advance material preparation and node assurance in final assembly pull scenarios, and reduces assembly waiting time and cycle time fluctuations.
Smart Images
Figure CN122433985A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent manufacturing and assembly material supply technology, and in particular to a method for predicting the shortage of key standard parts in a final assembly pull production scenario. Background Technology
[0002] As the manufacturing industry shifts towards multi-variety, small-batch, and flexible production methods, demand-driven final assembly production organization is gradually becoming an important organizational method for complex equipment manufacturing. In the final assembly of complex equipment such as aero-engines, assembly tasks typically revolve around orders and key assembly milestones. Whether materials can be readily assembled according to these milestones directly affects the start of assembly, assembly cycle time, and the progress of subsequent tasks. Especially in final assembly-driven scenarios, the frequent changes in material status place higher demands on the real-time performance and accuracy of material support during the assembly process.
[0003] During final assembly, critical standard components, while having relatively low individual value, are characterized by large usage, frequent consumption, and wide coverage, making them crucial materials affecting the continuous progress of assembly milestones. Especially during processes such as two-stage assembly and testing, disassembly and repair, and reassembly and retesting, the continued usability of critical standard components like self-locking nuts, locking plates, and rubber rings often depends on their condition after disassembly and reassembly. Influenced by factors such as disassembly and reuse, changes in quality status, and on-site replacement, the actual demand for critical standard components does not consistently follow the initial manufacturing bill of materials but is prone to significant fluctuations, leading to gaps in assembly milestones.
[0004] Current methods for managing the completeness of critical standard parts largely rely on static ledgers, inventory checks, and manual judgment. Gap identification typically occurs close to deadlines or even after parts shortages have already occurred, lacking the ability to predict future gaps. Without effective forecasting methods, critical standard part gaps often require reactive resupply and temporary coordination, increasing on-site support pressure and potentially causing assembly delays, cycle time fluctuations, and schedule adjustments. This fails to meet the requirements of advance material preparation, risk warning, and deadline assurance in final assembly pull scenarios. Therefore, there is an urgent need to propose a critical standard part completeness gap prediction method based on final assembly pull scenarios to enable early identification of future critical standard part gap risks and provide support for on-site material preparation and assembly assurance. Summary of the Invention
[0005] The purpose of this invention is to provide a method for predicting the kitting gap of key standard parts based on final assembly pull. By uniformly processing order tasks, assembly node plans, manufacturing bill of materials and key standard parts status data, a key standard parts gap prediction model is constructed to achieve early identification and warning of future gaps of key standard parts at target assembly nodes. This solves the problems of difficulty in accurately depicting the demand fluctuations of key standard parts, difficulty in timely identification of gap risks and delay in kitting warning in the existing technology.
[0006] The technical solution to achieve the purpose of this invention is as follows:
[0007] A method for predicting the kitting gap of key standard parts based on final assembly pull, characterized by the following steps:
[0008] Step 1: Obtain raw business data, which includes order task data, assembly node plan data, manufacturing bill of materials (MBOM) data, and key standard parts status data; where MBOM stands for Manufacturing Bill of Materials.
[0009] Step 2: Preprocess the original business data to establish the mapping relationship between key standard part identifiers, material items and assembly nodes, and obtain a standardized dataset.
[0010] Step 3: Based on the MBOM data, analyze the required quantity of key standard parts for the target assembly node, and calculate the effective available quantity and the shortage quantity of the target assembly node in combination with the status data of key standard parts.
[0011] Step 4: Construct the input feature vector for predicting the critical standard component gap, and input the input feature vector into the LSTM model to obtain the baseline prediction result of the critical standard component gap at future time; where LSTM stands for Long Short-Term Memory.
[0012] Step 5: Based on the prediction residual between the actual gap and the baseline prediction result, perform residual correction using the XGBoost model. XGBoost stands for Extreme Gradient Boosting.
[0013] Step 6: Combine the baseline prediction result with the residual correction result to obtain the future critical standard component gap prediction result. The combination method is to sum the baseline prediction result and the residual correction result and obtain the final prediction result through non-negative constraint processing.
[0014] Step 7: Output the gap list and kitting risk warning results for the target assembly node based on the gap prediction results of the key standard parts.
[0015] Furthermore, the preprocessing in step 2 includes: deduplicating duplicate state records, retaining conflicting state records according to the principle of prioritizing unavailable states, and completing missing numerical fields.
[0016] Furthermore, the number of material item gaps at the target assembly node in step 3 is calculated as follows: ,in, The quantity of this material item required by the target assembly node. The effective available quantity of this material item for the target assembly node.
[0017] Furthermore, the input feature vector in step 4 includes at least the demand, available quantity, quantity in transit, quantity awaiting inspection, quantity released, frozen quantity, quantity delivered, quantity occupied, supply cycle characteristics, node type characteristics, and the gap quantity at the previous moment.
[0018] Furthermore, in steps 4 to 6, the temporal features of the state changes of key standard parts are extracted using the LSTM model, the prediction residuals output by the LSTM model are fitted and corrected using the XGBoost model, and the prediction results of the key standard parts notches are obtained by combining the outputs.
[0019] The significant advantages of this invention compared to existing technologies are:
[0020] (1) This invention is geared towards the final assembly pull production scenario. It unifies and associates order tasks, assembly nodes, MBOM and dynamic status data of key standard parts, which can more accurately reflect the real guarantee status of key standard parts at the target assembly node.
[0021] (2) This invention comprehensively considers status information such as in transit, pending inspection, release, frozen, delivery and occupation, and extends the traditional static kitting judgment to gap prediction for future nodes, which is conducive to the early identification of gap risks of key standard parts.
[0022] (3) The present invention adopts a combination of LSTM and XGBoost prediction method, which takes into account the nonlinear relationship between the temporal and structural characteristics of the gap change of key standard parts, improves the accuracy and stability of the prediction of the gap of key standard parts, and can provide support for advance material preparation, complete set early warning and node guarantee in the general assembly pull scenario. Attached Figure Description
[0023] Figure 1 This invention proposes a method for predicting the missing parts of key standard components based on the final assembly pull. Detailed Implementation
[0024] This embodiment of a method for predicting the kitting gap of key standard parts based on final assembly pull includes the following steps:
[0025] Step 1: Obtain raw business data, including order task data, assembly node plan data, Manufacturing Bill of Materials (MBOM) data, and key standard component status data. The key standard component status data includes statuses such as in transit, in stock, awaiting inspection, released, frozen, delivered, and occupied. This data can be sourced from the company's existing business systems and on-site business records, and is used to reflect the demand formation process, status changes, and actual support status of key standard components during the final assembly stage.
[0026] Step 2: Preprocess the raw business data obtained in Step 1 to establish a mapping relationship between key standard component identifiers, material items, and assembly nodes, resulting in a standardized dataset. Specifically, firstly, field unification and association alignment are performed on the multi-source data. Since the MBOM requirement data and key standard component status data differ in data structure and record granularity, a mapping relationship of "key standard component identifier / batch identifier—material item—assembly node" is used to ensure that status data can be accurately backfilled into the requirement list of the corresponding assembly node. Secondly, duplicate status records, timestamp-reversed records, and conflicts involving multiple status records existing at the same time are handled. Within a preset time window, only the latest record of the same status is retained; when multiple conflicting statuses exist at the same time, unavailable statuses are retained first to avoid misclassifying key standard components that are actually unusable for assembly as available.
[0027] Furthermore, this embodiment uniformly maps the business status of critical standard parts into two categories: available and unavailable. Specifically, critical standard parts in stock that have been released and are not frozen, scrapped, or occupied by other tasks, as well as critical standard parts that have been delivered and meet assembly conditions, are classified as available. Critical standard parts that are in transit, awaiting inspection, frozen, scrapped, or occupied by other tasks are classified as unusable in the current assembly node. For missing values in key numerical fields such as required quantity, available quantity, and in-transit quantity, they are first filled in according to business consistency rules; if they cannot be filled in according to business rules, interpolation is then used for filling.
[0028] To reduce the impact of differences in feature dimensions on model training, this embodiment normalizes the input numerical features. The normalization formula is as follows:
[0029]
[0030] in, These are the normalized eigenvalues. and Features The maximum and minimum values in the sample set.
[0031] Step 3: Calculate the critical standard parts gap for the assembly node. Based on the normalized dataset obtained in Step 2, analyze the required quantity of critical standard parts for the target assembly node, and calculate the effective available quantity and gap quantity of the target assembly node in combination with the critical standard parts status data.
[0032] Set order At the assembly node The set of key standard component requirements is as follows:
[0033]
[0034] in, This indicates the key standard component material item. Indicates order At the assembly node For key standard parts material items The required quantity. The corresponding effective available quantity can be obtained from real-time status data. The number of gaps in the critical standard component material item at this assembly node is defined as follows:
[0035]
[0036] when When, it indicates that the key standard component is in the order. Assembly nodes There is a gap; when When this is the case, it indicates that the key standard component meets the requirements of the current assembly node.
[0037] Step 4: Input Feature Construction and Sample Generation
[0038] After completing the gap calculation, the input feature vector for predicting the gaps in key standard components is constructed. Let time be... The input feature vector is:
[0039]
[0040] in, Indicates the quantity demanded. Indicates the effective available quantity. Indicates the quantity in transit. Indicates the quantity to be inspected. Indicates the number of vehicles allowed. Indicates the quantity frozen. Indicates the quantity delivered. Indicates the quantity occupied. Indicates supply cycle characteristics, Indicates node type characteristics, This indicates the number of gaps at the previous moment.
[0041] In this embodiment, the sample set is divided into a training set, a validation set, and a test set according to time sequence to ensure that model training and testing are more consistent with actual business scenarios. Data from earlier periods is used as the training set, data from intermediate periods as the validation set, and data from later periods as the test set.
[0042] Step 5: Perform baseline prediction based on the LSTM model. Input the input feature vector obtained in Step 4 into the LSTM model to obtain the baseline prediction results of the critical standard parts shortage at future time. Since the status of critical standard parts, such as in transit, awaiting inspection, release, frozen, delivered, and occupied, changes continuously over time, the shortage has obvious temporal correlation. Therefore, the LSTM model is used to learn the temporal characteristics of the critical standard parts shortage changes.
[0043] In this embodiment, the LSTM model can be configured with the following parameters: 64 hidden units, a learning rate of 0.001, a batch size of 64, and a maximum number of iterations of 200. After training, the model outputs a baseline prediction value for the future gap in key standard components.
[0044] Step 6: Perform residual correction based on the XGBoost model. Based on the predicted residuals between the actual gap and the baseline prediction results obtained in Step 5, perform residual correction using the XGBoost model. Since the formation of critical standard component gaps is affected not only by temporal variations but also by structural features such as material properties, node types, and supply cycles, the XGBoost model is used to fit the predicted residuals output by the LSTM model to compensate for the nonlinear bias caused by structural features.
[0045] Let the actual gap be The baseline prediction result output by the LSTM model is The predicted residual is:
[0046]
[0047] Then, the XGBoost model is used to learn the residuals to obtain the residual correction values. In this embodiment, the XGBoost model can be configured with the following parameters: maximum tree depth of 20, learning rate of 0.05, number of weak learners of 300, subsample ratio of 0.8, and minimum split loss of 0.
[0048] Step 7: Combine the baseline prediction result obtained in Step 5 with the residual correction result obtained in Step 6 to obtain the future critical standard component gap prediction result, specifically expressed as follows:
[0049]
[0050] By performing the above non-negative truncation process, we can avoid producing negative gap prediction results that do not conform to physical meaning.
[0051] Furthermore, based on the predicted shortage results of key standard components, a shortage list for the target assembly node and a kitting risk warning result are output. The output results include at least the key standard component material items, corresponding assembly nodes, predicted shortage quantities, and warning information, which are used to support advance preparation of key standard components, assembly node assurance, and on-site plan adjustments.
[0052] On a set of test data, the method described in this embodiment was compared with the LSTM model and the XGBoost model alone. The results show that the combined model described in this embodiment has a mean absolute error (MAE) of 3.716, a root mean square error (RMSE) of 6.033, and a coefficient of determination of 0.858. The prediction performance is better than that of the single LSTM model and the single XGBoost model. This indicates that the method described in this embodiment can better identify the risk of shortages in key standard parts in advance and provide support for material preparation and assurance in the final assembly pull scenario.
Claims
1. A method for predicting the kitting gap of key standard parts based on final assembly pull, characterized in that, Includes the following steps: Step 1: Obtain raw business data, which includes order task data, assembly node plan data, manufacturing bill of materials (MBOM) data, and key standard parts status data. Step 2: Preprocess the original business data to establish the mapping relationship between key standard part identifiers, material items and assembly nodes, and obtain a standardized dataset; Step 3: Based on the Manufacturing Bill of Materials (MBOM) data, analyze the required quantity of key standard parts for the target assembly node, and calculate the effective available quantity and the shortage quantity of the target assembly node in combination with the status data of key standard parts; Step 4: Construct the input feature vector for predicting the gaps in key standard components, and input the input feature vector into the LSTM model to obtain the baseline prediction result of the gaps in key standard components at future time. Step 5: Based on the prediction residual between the actual gap and the baseline prediction result, perform residual correction using the XGBoost model; Step 6: Combine the baseline prediction results with the residual correction results to obtain the future key standard component gap prediction results; Step 7: Output the gap list and kitting risk warning results for the target assembly node based on the gap prediction results of the key standard parts.
2. The method for predicting the kitting gap of key standard parts based on final assembly pull according to claim 1, characterized in that, The preprocessing described in step 2 includes: deduplicating duplicate state records, retaining conflicting state records according to the principle of prioritizing unavailable states, and completing missing numerical fields.
3. The method for predicting the kitting gap of key standard parts based on final assembly pull according to claim 1, characterized in that, In step 3, the number of material item gaps at the target assembly node is calculated as follows: ,in, The quantity of this material item required by the target assembly node. The effective available quantity of this material item for the target assembly node.
4. The method for predicting the kitting gap of key standard parts based on final assembly pull according to claim 1, characterized in that, The input feature vector in step 4 includes demand, available quantity, quantity in transit, quantity awaiting inspection, quantity released, frozen quantity, quantity delivered, quantity occupied, supply cycle characteristics, node type characteristics, and the gap quantity at the previous moment.
5. The method for predicting the kitting gap of key standard parts based on final assembly pull according to claim 1, characterized in that, In steps 5 and 6, the temporal features of the state changes of key standard parts are extracted using the LSTM model, the prediction residuals output by the LSTM model are fitted and corrected using the XGBoost model, and the prediction results of the key standard parts notches are output using a non-negative truncation method.
6. The method for predicting the kitting gap of key standard parts based on final assembly pull, as described in claim 1, is characterized in that... The kitting risk warning results mentioned in step 7 include at least the number of missing items, the missing material items, and the warning information.