From Materials Discovery to Deployment: A Critical Review of Artificial Intelligence Methods Across the Sensor Development Pipeline

Sensors are the silent sentinels of the modern world, the tiny devices that measure the temperature of a patient's skin, the pressure of a car's tire, or the presence of a toxic gas in the air. For decades, creating these devices has been a slow, laborious craft. Engineers would discover a new material, design a physical structure around it, build a prototype, and then spend years tweaking the recipe and calibrating the device until it worked reliably. This entire journey, from a spark of an idea to a product on a shelf, could easily take twenty years. Recently, a new force has entered the workshop: artificial intelligence. The promise is that these smart algorithms can not only make sensors work better but can also shrink the entire twenty-year development cycle down to just a few years. This idea has generated immense excitement, suggesting that machines might soon design and build the sensors of the future entirely on their own.

A new critical review by researchers at Dayananda Sagar University in India examines this promise with a sharp, skeptical eye. The authors did not simply list every time someone used a computer to help with a sensor; instead, they mapped out the entire lifecycle of sensor creation, from the initial discovery of materials to the final deployment in the field. They asked a simple but vital question: for each step of this process, is the speed-up real and proven, or is it just a hopeful guess? By sorting the evidence into three categories—what is already being used in factories, what has been shown in a laboratory but not yet scaled up, and what exists only as a theoretical idea—they found that the reality is far more nuanced than the headlines suggest.

The review reveals that artificial intelligence is already a mature, working tool in specific, well-defined corners of sensor engineering. In the final stages of production, where sensors are built on silicon wafers, computer vision systems are routinely used to spot microscopic defects. These systems are so reliable that they are considered a standard part of the manufacturing process in the semiconductor industry. Similarly, once a sensor is built, artificial intelligence is proving highly effective at calibration. Sensors often drift over time, giving slightly different readings as they age or as the environment changes. Algorithms can now learn from a few reference sensors and adjust the rest of a large group, cutting the time needed to calibrate them by up to 93 percent. In these areas, the technology is not a future dream; it is a present-day reality that is already saving time and money.

However, the story changes when the researchers look at the earlier, more creative stages of development, such as discovering new materials or designing the physical shape of the sensor. Here, the evidence is much thinner. While computers can predict the properties of new materials with impressive speed, the review found that these predictions rarely translate into actual, working sensors that have been built and tested. There are examples of "self-driving laboratories," where robots mix chemicals and test them without human intervention, but these successes have mostly been limited to specific types of chemistry and have not yet been proven to work for the wide variety of sensors used in the real world. The authors note that while these robotic labs are real and peer-reviewed, they are currently more like sophisticated experiments than universal factories.

Perhaps the most striking finding concerns the grandest claim of all: that artificial intelligence can compress the entire twenty-year development pipeline into a few years. The authors argue that this specific vision is currently a projection, not a proven fact. It is a concept drawn from the most data-rich fields of science, where computers have vast amounts of information to learn from. In the world of sensors, data is often scarce, expensive to generate, and proprietary. Because of this lack of data, the idea of a fully autonomous pipeline that designs, builds, and tests a sensor from start to finish remains a theoretical possibility rather than an industrial reality. The review highlights that while we have powerful tools for specific tasks, we do not yet have a single, seamless system that can replace the decades-long human effort across all stages of sensor creation.

The researchers also point out a significant gap in the field: the lack of shared data and standard benchmarks. In many scientific disciplines, researchers share their results openly so others can verify them. In sensor development, data is often locked away in private companies or lost because failed experiments are rarely published. This secrecy makes it difficult to know if a new artificial intelligence method is truly better or just lucky with one specific dataset. The authors suggest that the field would advance faster if scientists focused less on inventing new algorithms and more on building shared databases and testing standards. Without these foundations, the promise of a fully automated future for sensors will remain just that—a promise.

Ultimately, this review serves as a necessary reality check for a field buzzing with optimism. It confirms that artificial intelligence is a powerful partner in sensor engineering, but it is not a magic wand that solves every problem instantly. The technology is ready for the job in areas where the rules are clear and data is plentiful, such as checking for defects or adjusting sensor readings. But for the creative, messy work of inventing new materials and designing complex devices, the technology is still in its infancy. The path forward, the authors conclude, is not to expect a sudden revolution, but to build the shared infrastructure that will allow these smart tools to finally prove their worth across the entire development pipeline.