Deep Learning-Assisted Multi-Objective Optimization and Microstructure Property Correlation in Rotary Friction Welding of Dissimilar AA1100–AA2014 Aluminium Alloys

Imagine you are trying to bake the perfect cake, but you have two very different ingredients: one is a soft, stretchy marshmallow, and the other is a hard, crunchy cookie. If you try to melt them together in a normal oven, the marshmallow might turn into a sticky mess while the cookie stays hard, or they might separate completely. This is exactly the problem engineers face when trying to join different types of aluminum. One type is soft and great at conducting heat (like the marshmallow), while the other is super strong but tricky to work with (like the cookie). Usually, when you try to weld them using high heat, they don't mix well, leaving weak spots that could break under pressure.

To solve this, scientists use a technique called "Rotary Friction Welding." Instead of melting the metal like a normal weld, they spin one piece against the other at high speed. Think of it like rubbing your hands together really fast to make them hot; the friction creates enough heat to make the metal soft and squishy without actually melting it. Then, they smash them together with a heavy hammer (called forging) to fuse them into one solid piece. The tricky part is figuring out exactly how fast to spin, how hard to press, and for how long. If you get the recipe wrong, the joint is weak. This is where the paper steps in, using a mix of old-school experiments and super-smart computer brains to find the perfect recipe for joining these two very different aluminum alloys.

The Great Aluminum Match-Up

In this study, researchers tackled the challenge of joining two specific aluminum alloys: AA1100 and AA2014. You can think of AA1100 as the "gentle giant"—it's very soft, stretches easily, and conducts heat like a champ. On the other hand, AA2014 is the "tough brawler"—it's incredibly strong, hard, and built to carry heavy loads, but it's a bit more brittle. Joining them is like trying to glue a rubber band to a steel rod; if you do it wrong, the bond fails.

The team used a method called Rotary Friction Welding (RFW). Imagine holding two pencils, spinning one against the other until they get hot and soft, then pressing them together. In the factory, this is done with metal cylinders. The researchers had five "knobs" to turn to control this process: how fast the metal spins (RPM), how hard they press it while spinning (friction pressure), how hard they press it after stopping (forging pressure), and how long they do each step (time).

The Experiment: A Recipe Search

To find the perfect settings, the researchers didn't just guess. They used a systematic plan called a Taguchi L27 design. This is like a chef testing 27 different variations of a recipe, changing the amount of sugar, flour, and baking time in a specific pattern to see which combination makes the best cake. They ran 27 different welding experiments, changing the speed, pressure, and time for each one.

After welding, they put the joints through the wringer. They pulled them apart to see how strong they were (tensile strength), measured how hard the surface was (microhardness), and stretched them to see how much they could bend before breaking (elongation). They also looked at the metal under powerful microscopes (SEM) and used X-rays (XRD) to see what the metal looked like on a tiny, crystal level.

The Computer Brain: Deep Learning

Here is where the study gets really futuristic. Instead of just looking at the 27 results and guessing the pattern, the researchers built a Deep Neural Network (DNN). Think of this as a digital brain that learns by example. They fed the computer all the data from the 27 experiments—the settings they used and the results they got. The computer's job was to figure out the hidden, complicated rules that connect the settings to the strength of the weld.

Once the computer learned the rules, they used a second smart tool called NSGA-II. This is like a super-optimizer that searches for the "perfect" balance. Since you can't always have maximum strength and maximum stretchiness at the same time (they often fight each other), the optimizer finds the best possible compromise, known as a "Pareto front." It's like finding the perfect spot on a map where you are close to both the beach and the mountains, rather than being stuck in the middle of nowhere.

What They Found

The results were impressive. The "digital brain" learned the rules incredibly well. When the researchers tested the computer's predictions against real-world results, the computer was right about 98.7% of the time (an R² value of 0.987). This means the computer model is a very reliable tool for predicting how the weld will behave without needing to build a physical sample every time.

The study found that the best welding recipe involved spinning at 1200 RPM, with a friction pressure of 70 MPa, a forging pressure of 90 MPa, a friction time of 4 seconds, and a forging time of 5 seconds.

Using these settings, they created a joint that was:

• 142 MPa strong (a significant jump from the weakest joints they tested, which were only 95 MPa).
• 122.3 HV hard.
• Able to stretch 10% before breaking.

The Secret Sauce: What's Happening Inside?

Why did these settings work so well? The researchers looked under the microscope and found the answer. In the best joints, the metal grains (the tiny crystals that make up the metal) became very fine and uniform, like a smooth, tightly packed crowd. This happened because the heat and pressure caused a process called dynamic recrystallization, where the metal reorganizes itself into a stronger structure.

Crucially, they found no harmful cracks or weird chemical reactions (intermetallic phases) at the join. The X-ray analysis showed that the metal kept its original structure without turning into brittle, weak compounds. The AA1100 side showed a "ductile fracture," meaning it stretched and tore like a soft plastic bag, while the AA2014 side showed signs of its hard, precipitate-filled structure. Despite these differences, the two metals fused together perfectly, creating a seamless bond.

Why This Matters

This paper suggests that by combining real-world experiments with a smart computer model, engineers can quickly find the best way to join different metals without wasting time and money on endless trial-and-error. The "hybrid" approach—using a Deep Neural Network to predict results and an optimizer to find the best settings—acts like a high-tech decision-making tool.

The researchers believe this method could be a game-changer for industries like aerospace, automotive, and railway, where making lightweight but strong structures is essential. By mastering the art of joining dissimilar aluminum alloys, they can build lighter planes and cars that use less fuel, all while ensuring the joints are strong enough to keep everyone safe. The study confirms that with the right "recipe" and a little help from artificial intelligence, even the most mismatched materials can become a perfect team.