TL;DR
This paper introduces a novel method to accurately infer the global residual stress field (RSF) in complex die-forged aircraft parts. By fusing finite element analysis (FEA) simulation data (as a prior) with real-time machining monitoring data, the approach overcomes the limitations of traditional methods, enabling better process control and improved component reliability.
What Problem Does This Paper Address?
Die-forged structural parts are critical in aircraft due to their excellent mechanical properties. However, the complex forging and heat treatment processes induce significant and intricate residual stress fields (RSFs) within these parts. These RSFs can negatively impact deformation, fatigue life, and overall component reliability. Accurately understanding the global RSF is crucial for effective process control. Existing methods for inferring RSFs often struggle with the irregular geometries and complex stress distributions of die-forged parts, leading to ill-posed problems and inaccurate results. Traditional measurement techniques are often local or destructive, making it difficult to capture the complete RSF.
Methods
The proposed method combines two key components:
- Prior Acquisition of RSF Distribution: This step leverages FEA simulations of the die-forging and heat treatment processes. The simulation provides a "prior" distribution trend of the RSF within the part. This prior knowledge is crucial for reducing the complexity of the problem.
- Low-Dimensional RSF Characterization and Inference:
- Characterization: The RSF is characterized in a low-dimensional manner by grouping elements with similar stress levels. This is achieved by dividing the stress range for each direction (e.g., $\sigma_{11}$) into $n$ groups. Within each group, the stress distribution is then approximated by a simple linear function, significantly reducing the number of parameters needed to represent the RSF (from thousands of element stresses to $2 \times n \times 6$ parameters for six stress directions).
- Inference: A deformation-force-based inference method is employed. The relationship between the deformation force (measured during machining) and the RSF is established using the principle of virtual work. To handle uncertainties and noise in the deformation force data (common in real-world machining), a Bayesian inference framework is utilized. This framework treats the RSF parameters as random variables and updates prior beliefs with observed data to obtain a stable and reliable posterior estimate of the RSF.
Datasets
The paper utilizes:
- Simulated Data: FEA simulations (using DEFORM software) of the die-forging and heat treatment processes for a 7050 aluminum alloy part were used to generate the prior RSF distribution.
- Simulated Machining Data: The machining process of the part was simulated (using ABAQUS software) to generate deformation force data. This simulated data was then intentionally corrupted with Gaussian noise to mimic real-world measurement uncertainties.
- Actual Machining Data: Deformation force data was collected from actual machining experiments on die-forged parts using intelligent fixtures with force sensors. This data was used for validation.
No new benchmark datasets were introduced; rather, existing simulation and experimental setups were used for evaluation.
Key Insights and Ideas
- Prior Knowledge is Powerful: Using FEA-derived RSF distribution trends as a "prior" is a highly effective strategy for tackling the ill-posed inverse problem of RSF inference, especially for complex geometries.
- Dimensionality Reduction is Key: Characterizing the RSF by grouping elements and fitting linear functions dramatically reduces the number of parameters, making the inference computationally feasible and more robust.
- Bayesian Inference for Robustness: The Bayesian framework is essential for handling the inherent noise and uncertainties in real-world deformation force measurements, leading to more reliable RSF estimates.
- Deformation Force as a Proxy: The deformation force measured during machining serves as a valuable indirect indicator of the underlying RSF, bridging the gap between manufacturing processes and internal stress states.
Key Results
- Simulation Verification: The proposed method achieved accurate RSF inference in a simulated environment, with average absolute errors of 0.54 for parameter 'a' and 0.382 for parameter 'b'. The predicted deformation force also showed a low average absolute error of 0.294 N.
- Actual Machining Verification: In actual machining, the method demonstrated its effectiveness. The predicted deformation force for the eighth layer of machining had an average absolute error of 121.4 N (15.1% relative error).
- Impact on Deformation Prediction: The RSF inference error introduced an error of approximately 13.7% in subsequent machining deformation predictions. Despite this, the average absolute error at the corner points of the part was less than 0.1 mm, indicating the method's practical utility.
Limitations and Future Work
- Accuracy of Prior: The accuracy of the RSF inference is influenced by the difference between the theoretical (simulated) and actual RSF distributions.
- Partitioning and Function Choice: The number and division strategy of RSF groups, as well as the choice of characterization functions, can impact accuracy.
- Future Work: The authors suggest incorporating more residual stress measurement methods (like drilling and ultrasonic) to improve the prior acquisition. Exploring different partitioning strategies and characterization functions could further enhance the global RSF inference accuracy. The method's applicability to other materials (e.g., titanium alloys) and linear elastic materials is also highlighted as a future direction.
Summary updated