Method for predicting connecting rod finished product weight using ai technology

Abstract

This invention provides a method for predicting the weight of finished connecting rods using AI technology, belonging to the field of engine manufacturing technology. Addressing the technical problem of the lack of a clear linear relationship between the weight of diesel engine connecting rod blanks and the finished product, and the inability of traditional methods to accurately predict the quality of the finished product weight at the blank stage, this invention constructs a training dataset by collecting key dimensional parameters and weights of the connecting rod blanks and their corresponding finished product weights. Using the key dimensional parameters and weight of the blanks as input features and the finished product weight as the output label, an AI prediction model based on gradient boosting trees or random forest algorithms is trained. The trained model is deployed to a detection terminal, where new blank parameters are input, and the predicted finished product weight is output in real time and compared with a preset acceptable range to quickly output the quality assessment result. This invention achieves accurate prediction of the finished product weight at the blank stage through a multi-dimensional feature system and nonlinear mapping modeling, significantly reducing production costs and improving production efficiency.

Description

Technical Field

[0001] This invention belongs to the field of engine manufacturing technology, specifically relating to a method for predicting the weight of finished connecting rods using AI technology. It is particularly suitable for accurately predicting the weight of finished products at the blank stage, thereby achieving pre-quality control. Background Technology

[0002] As a core transmission component of the diesel engine, the connecting rod's weight must be strictly controlled within the design range, directly affecting the engine's power output stability, operational balance, and service life. Taking a certain type of diesel engine connecting rod as an example... Figure 1-2 The simplified diagram of the connecting rod blank shown is as follows: Figure 3-4 The diagram shown is a simplified representation of the finished connecting rod. The weight tolerance of the finished product is extremely small, with the small end requiring a weight of 19.7-20.1 kg and the large end requiring a weight of 52.2-53 kg, for a total weight tolerance of only 1.2 kg.

[0003] In the existing manufacturing process, the weight control range of connecting rod blanks is 91-94.5 kg, with a tolerance of 3.5 kg, showing significant fluctuations. The dimensional tolerances of the blanks are also large (e.g., rod thickness, rod width, large and small end hole diameters, and center distance can reach ±2 mm), accounting for over 95% of the overall impact. Furthermore, some areas of the connecting rod blank surface (such as the surface of the rod thickness and width, and the surface of the rod cap thickness) are retained, while other areas (such as the large and small end holes and their end faces, bolt holes and bolt positioning surfaces, and the rough surfaces of the rod body and rod cap) need to be removed by machining. Because there is no clear linear relationship between the blank and the finished product, traditional methods rely solely on whether the blank weight is within the theoretical range to predict whether the finished product is qualified, which has the following drawbacks: 1. Low prediction accuracy: The blank surface is partially preserved and partially processed away, which makes it impossible to establish a clear linear relationship between the weight of the blank and the finished product, and thus makes it impossible to accurately predict the weight of the finished product; 2. Serious waste of resources: Unqualified blanks must be processed completely before they are rejected, resulting in multiple wastes of raw materials, processing time, energy and other resources; 3. Low production efficiency: Finished product weight inspection needs to be carried out after processing, which has a long judgment cycle and affects production nodes. If qualified parts are missing, it will lead to delays in the assembly of the whole machine. 4. Lack of pre-control: It is impossible to screen out defective products at the raw material stage, and quality control is in a post-inspection mode, which cannot achieve lean production and does not conform to the development direction of intelligent manufacturing.

[0004] Therefore, there is an urgent need for a method that can accurately predict the weight of the finished connecting rod at the blank stage. Summary of the Invention

[0005] The technical problem solved by this invention: This invention aims to solve the technical problems of low accuracy in predicting the weight of finished connecting rods and the inability to determine their conformity at the blank stage in the prior art. It provides a method for predicting the weight of finished connecting rods using AI technology, so as to achieve rapid and accurate prediction of the weight of finished products at the blank stage, thereby reducing production costs, improving production efficiency, and promoting the transformation of connecting rod manufacturing towards lean and intelligent manufacturing.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for predicting the weight of a finished connecting rod using AI technology includes the following steps: (1) Collect key dimensional parameters, blank weight, and corresponding finished product weight data of the connecting rod blank to construct a training dataset. In this step, the finished product weight data can be obtained by actually weighing the finished product using an electronic scale; (2) Train the AI ​​prediction model with the key dimension parameters and weight of the blank as input features and the weight of the finished product as the output label; (3) Deploy the trained model to the detection terminal, input the new blank parameters, and output the predicted weight of the finished product; (4) Based on the comparison between the predicted value and the preset acceptable range, output the finished product's acceptance result. In this step, the acceptable range is determined based on the weight range specified in the drawings, with reference to the theoretical weight measured by 3D modeling; In step (1), the key dimensional parameters include one or more of the following: shaft thickness, shaft width, diameter of the large and small end holes, center distance between the large and small end holes, and radius of the rounded corner.

[0007] In step (2), the AI ​​prediction model is either the Gradient Boosting Tree (XGBoost) or the Random Forest algorithm.

[0008] Step (1) also includes outlier removal, missing value filling and standardization preprocessing of the data.

[0009] In step (2), the hyperparameters of the model are optimized by grid search, and cross-validation is introduced to prevent overfitting.

[0010] The hyperparameters mentioned in step (2) include one or more of the following: decision tree depth, learning rate, and node splitting threshold.

[0011] The detection terminal in step (3) supports manual input or automatic acquisition of blank parameters by the sensor.

[0012] Step (4) also includes generating a test report and recording the predicted value and the judgment result.

[0013] It also includes periodically collecting new production data to incrementally train the model in order to adapt to fluctuations in production processes.

[0014] The fluctuations in the production process include one or more of the following: batch changes in materials, mold wear.

[0015] Advantages of this invention compared to existing technologies: 1. The prediction accuracy of this solution is significantly improved: through multi-dimensional feature fusion and nonlinear modeling, the prediction error of the finished product weight is controlled within ±0.3Kg, meeting stringent engineering requirements; 2. This solution achieves pre-process quality control: the quality of finished products can be determined at the blank stage, preventing unqualified blanks from entering subsequent processing, thus truly achieving pre-process control; 3. This solution significantly reduces production costs: it reduces raw material waste, processing time losses, and energy consumption, resulting in an overall production cost reduction of over 15%. 4. This solution improves production efficiency: shortens the production cycle, ensures assembly milestones, and improves production line response speed; 5. This solution has a wide range of applications: it can be extended to different models of connecting rods and other similar structural components, and can be quickly adapted by simply supplementing training data; 6. This solution's model has strong dynamic iteration capabilities: it supports incremental learning, adapts to process changes, and ensures long-term stable operation; 7. This solution is easy to operate: After deployment, it is simple to operate and can be used by non-professionals, making it suitable for industrial promotion. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the main structure of the connecting rod blank in this invention; Figure 2 This is a top view of the connecting rod blank structure in this invention; Figure 3 This is a schematic diagram of the main structure of the finished connecting rod in this invention; Figure 4 This is a top view of the finished connecting rod structure in this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1-4 The embodiments of the present invention are described in detail below.

[0019] This invention provides a method for predicting the weight of a finished connecting rod using AI technology, specifically including the following steps: Step (1): Data Acquisition and Preprocessing Collect at least 200 sets of historical production data, including key dimensions of the blank, blank weight, finished product weight, and qualification results; perform outlier removal, missing value filling, and standardization on the data to construct training and testing sets; the key dimensions include rod thickness, rod width, diameter of the large and small end holes, center distance between the large and small end holes, and radius of the fillet.

[0020] Step (2): AI Model Construction and Training Using key dimensions and weight of the blank as input features, and finished product weight as the output label, a prediction model based on either XGBoost or Random Forest algorithms is constructed. Hyperparameters (such as decision tree depth, learning rate, and node splitting threshold) are optimized through grid search to improve the model's generalization ability. Cross-validation (such as 5-fold cross-validation) is introduced during training to prevent overfitting and ensure prediction stability across different batches of blank data.

[0021] Step (3): Model Deployment and Judgment The trained model is deployed to an industrial inspection terminal (supports PC and industrial tablet). It supports manual input or automatic acquisition of blank parameters by sensors. The model outputs the predicted weight of the finished product in real time, compares it with the preset qualified range, outputs the qualified judgment result, and generates an inspection report.

[0022] Step (4): Model Iteration and Optimization Regularly collect new production data (e.g., monthly) to incrementally train the model, continuously optimize model parameters, adapt to fluctuations in the blank production process (e.g., material batch changes, mold wear, etc.), and ensure long-term prediction accuracy.

[0023] An embodiment of the present invention: Step (1) Data Acquisition and Model Building: Taking a certain type of diesel engine connecting rod as an example, 350 sets of historical production data were collected, covering three production batches. Ten key dimensional parameters of the blank, including rod thickness, rod width, diameter of the large and small ends, and center distance, were measured using a coordinate measuring machine (or manual measurement, or 3D scanning measurement). The weight of the blank and the corresponding finished product were weighed using an electronic scale with an accuracy of 0.01 kg. After outlier removal (e.g., removing extreme data caused by measurement errors) and standardization, a training set (280 sets) and a test set (70 sets) were constructed. The measurement method using a coordinate measuring machine in this step is as follows: the machining reference surface of the connecting rod faces the measuring head, using the machining reference surface as the measurement reference surface, supplemented by the large and small ends as the measurement reference, so that the measurement reference is consistent with the machining reference, reducing measurement errors. The coordinate measuring machine mainly measures the center distance and diameter of the connecting rod's large and small ends, and the average value of three measurements is taken as the measured value of the large and small ends. For parts such as the thickness and width of the rod that are easy to measure with ordinary measuring instruments, vernier calipers are used for measurement. The sampling point is divided into three sections along the length of the connecting rod, and the average value is taken as the measurement value after three measurements.

[0024] Step (2) AI Model Construction and Training: The prediction model is constructed using the XGBoost algorithm. The input features are 10-dimensional, and the output is the predicted weight of the finished product. Hyperparameters are optimized through grid search, with the decision tree depth set to 6, the learning rate to 0.1, and the node splitting threshold to 0.01. Five-fold cross-validation is introduced during training. The model's mean absolute error on the test set is 0.25 kg, and the maximum error does not exceed 0.4 kg, meeting production requirements.

[0025] Step (3) Model Deployment and Online Judgment The trained model is deployed to an industrial tablet terminal. After measuring the blank, the operator manually inputs the dimensional parameters and blank weight via a touchscreen. The model outputs a predicted finished product weight within 0.5 seconds. The system automatically compares this value with a preset acceptable range (19.7-20.1 kg for the smaller end and 52.2-53 kg for the larger end), displays a "qualified" or "unqualified" result, and generates an inspection report containing the predicted value, judgment conclusion, and measurement time. This process has been piloted on a connecting rod production line, with a cumulative accuracy of 96.5% for 500 blanks inspected.

[0026] In this step, in addition to the mapping relationship between the original dimensions and the finished product weight, new derived features are constructed by utilizing the relationship between measured and theoretical values ​​of various features. For example, the difference between the center distance of the blank and the theoretical center distance of the large and small end holes is used as a new feature, and the ratio of the measured to the theoretical thickness of the rod is used as a new feature. These subdivided features are used as part of the training data to improve the accuracy of the prediction results.

[0027] Step (4) Model Iteration and Process Adaptation Three months later, mold wear caused a shift in the size distribution of the blanks, and the model prediction error rose to 0.45 kg. By collecting 200 sets of data from a new batch of blanks and combining them with the original dataset for incremental model training, and adjusting the model parameters, the prediction error recovered to 0.28 kg. This example verifies the model's dynamic iterative capability, effectively adapting to changes in production processes.

[0028] Applying this method to another model of connecting rod only requires re-collecting 200 sets of historical data for that model and training the model using the same steps to achieve finished product weight prediction. Test results show that the prediction error is controlled within ±0.35Kg, proving that the method has good versatility and scalability.

[0029] This invention constructs a multi-dimensional feature system, integrating key geometric parameters of the connecting rod blank (such as rod thickness, width, bore diameter, center distance, etc.) and blank weight. It utilizes AI algorithms to automatically learn the complex nonlinear relationship between these features and the finished product weight. During training, the model establishes mapping patterns using extensive historical data. After deployment, it can quickly output predicted finished product weight values ​​at the blank stage, enabling pre-judgment of finished product quality. Its core lies in breaking through the traditional linear thinking of single blank weight prediction, capturing the interactions and nonlinear characteristics between multiple variables through machine learning algorithms, significantly improving prediction accuracy.

[0030] This invention combines AI with the manufacturing process of connecting rods to solve engineering problems. Essentially, it explores how data-driven intelligent prediction can be implemented in industrial processes in the physical world, with the core being the prediction of outcome quantities from process quantities. The principle and process are divided into three stages: the training stage is the parameterization from "experience" to "model"; the inference stage is the deduction from "blank" to "finished product," adhering to the fundamental laws of conservation of mass and volume, where blank volume - volume removed = finished product volume; and the feedback loop stage is the iterative learning from "error" to "optimization."

[0031] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0032] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for predicting the weight of a finished connecting rod using AI technology, characterized in that: Includes the following steps: (1) Collect key dimensional parameters, blank weight and corresponding finished product weight data of connecting rod blanks, and construct training dataset; (2) Train an AI prediction model with the key dimensions and weight of the blank as input features and the weight of the finished product as the output label; (3) Deploy the trained model to the detection terminal, input the new blank parameters, and output the predicted weight of the finished product; (4) Based on the comparison between the predicted value and the preset qualified range, output the qualified result of the finished product.

2. The method for predicting the weight of a finished connecting rod using AI technology according to claim 1, characterized in that: In step (1), the key dimensional parameters include one or more of the following: shaft thickness, shaft width, diameter of the large and small end holes, center distance between the large and small end holes, and radius of the rounded corner.

3. The method for predicting the weight of a finished connecting rod using AI technology according to claim 1, characterized in that: In step (2), the AI ​​prediction model is either Gradient Boosting Tree (XGBoost) or Random Forest algorithm.

4. The method for predicting the weight of a finished connecting rod using AI technology according to claim 1, characterized in that: Step (1) also includes outlier removal, missing value filling and standardization preprocessing of the data.

5. The method for predicting the weight of a finished connecting rod using AI technology according to claim 1, characterized in that: In step (2), the hyperparameters of the model are optimized by grid search, and cross-validation is introduced to prevent overfitting.

6. The method for predicting the weight of a finished connecting rod using AI technology according to claim 1, characterized in that: The hyperparameters mentioned in step (2) include one or more of the following: decision tree depth, learning rate, and node splitting threshold.

7. The method for predicting the weight of a finished connecting rod using AI technology according to claim 1, characterized in that: The detection terminal in step (3) supports manual input or automatic acquisition of blank parameters by the sensor.

8. The method for predicting the weight of a finished connecting rod using AI technology according to claim 1, characterized in that: Step (4) also includes generating a test report and recording the predicted value and the judgment result.

9. The method for predicting the weight of a finished connecting rod using AI technology according to claim 1, characterized in that: It also includes periodically collecting new production data to incrementally train the model in order to adapt to fluctuations in production processes.

10. The method for predicting the weight of a finished connecting rod using AI technology according to claim 9, characterized in that: The fluctuations in the production process include one or more of the following: batch changes in materials, mold wear.

Images