Liu Li fell asleep from working overtime for three consecutive days. When she woke up again, she was in a 1972 apartment building. Liu Li was faced with a major crisis: she was about to graduate fr...
The failure of the first bench test was like a bucket of ice water, chilling the three of them to the bone. The problem was clearly in the control system—the "brain" that was trying to "control" the vibration had malfunctioned.
Zhou Wei locked himself in his office, staring intently at the lines of complex code and mathematical models on his computer screen, his brow furrowed in deep thought. His previously designed control algorithm was based on classical linear control theory, aiming for precise mathematical models and optimal control parameters. But the reality was that the system they faced was fraught with uncertainty: coarse sensor signals carried noise and delay, friction and clearances in mechanical parts were nonlinear, and the dynamic characteristics under high-speed rotation were complex and variable…
He tried adjusting the parameters and reducing the control gain, hoping to make the system's response "softer." However, in the simulation, the response was either too slow to keep up with the changes in vibration, or the system became unstable again and repeated the same problem when the gain was increased slightly.
“No, it still doesn’t work…” Zhou Wei scratched his head in frustration. The simulation on the screen was interrupted again due to instability. “The nonlinearity of this system is too strong and the time delay is obvious. Using traditional pId (proportional-integral-derivative) control, the parameter tuning range is too narrow. We simply cannot find that ‘sweet spot’ that can effectively suppress vibration and maintain stability.”
He attempted to introduce more complex modern control theories, such as adaptive control, but this required online identification of system parameters, which involved a large amount of computation and placed high demands on the processor. With their existing rudimentary hardware, this was simply impossible.
The dilemma of algorithms is like an invisible mountain, blocking the way forward. Zhou Wei felt an unprecedented sense of frustration; theoretical knowledge seemed so powerless in the face of the complexity and harshness of the real world.
Liu Li and Sun Mei were also worried as they watched Zhou Wei grow increasingly haggard and anxious.
“Mr. Zhou, how about…we try a different approach?” Liu Li looked at Zhou Wei’s bloodshot eyes and tentatively suggested, “Don’t always try to use a perfect mathematical formula to ‘calculate’ everything. Could we…simulate how an experienced master craftsman does it?”
She recalled the scene when Master Wang Jianguo taught her how to grind drill bits and adjust machine tools: "Master Wang never adjusted the equipment by calculation. He listened to the sound, looked at the iron filings, felt the vibration, and then made small adjustments little by little based on his 'feel'. He didn't have complicated formulas in his mind, but he could find a 'point' that made the machine run smoothly. Can our device also learn from this kind of 'experience' adjustment?"
Zhou Wei paused for a moment, then shook his head with a wry smile: "Experience? Feel? That's too vague. How can you quantify it? How can you turn it into an algorithm?"
“It’s hard to quantify,” Liu Li admitted, “but the experienced technicians often grasp the main problem. For example, if they feel that the vibration is too strong, they will try to adjust it in a certain direction to see the effect. If it gets worse, they will adjust it back. If it gets better, they will adjust it a little more. Isn’t this a process of trial and error and finding the optimal solution? It’s just that the human brain can make judgments quickly, while our devices react slowly.”
Liu Li's simple idea, based on practical experience and bordering on "fuzzy control," contrasted sharply with Zhou Wei's pursuit of precise mathematical description. Zhou Wei instinctively felt it was not "scientific" enough and too "primitive," but the path to precise models seemed to have reached a dead end, forcing him to seriously consider this alternative possibility.
In the lab, the dilemma of algorithms still loomed over them. Should they continue to search for that faint hope within the framework of traditional theories, or boldly try a new path that is closer to human experience but whose future is uncertain? The choice was before them.