Complex production machines are critical to generating revenue, and they require continuous operation. The cost of machine downtime can be significant, potentially reaching hundreds of thousands of dollars per hour. This makes consistent maintenance necessary to help ensure reliability, quality and flawless functioning. Typically, machine manufacturers rely on routine maintenance services to avoid malfunctions and downtimes. However, these standard maintenance intervals are often labor-intensive and not always necessary, especially if the machinery is still functioning correctly.
Reducing Maintenance Intervals
Manufacturers rely on traditional maintenance services, but even the most qualified staff cannot detect all impending failures, leaving room for unexpected machine malfunctions.
"Recent advancements in the field of artificial intelligence have provided a solution to machine monitoring," said Viacheslav Gromov, CEO of the AI company AITAD. "Continuous sensor-based monitoring allows for the capture of vast amounts of data that an AI can analyze to reveal the machine's condition or individual components. The AI's evaluation is so in depth that it can even anticipate potential malfunctions well in advance. This innovative solution is beneficial for both manufacturers and users, as it ensures operational continuity at lower maintenance costs."
Nevertheless, most contemporary AI solutions require significant computational power and an expensive, high-capacity network infrastructure. After the sensor data is collected, it is transmitted to a central server, where the data can be evaluated before it is finally returned to the machine. This process is not only resource-inefficient, but also raises data security concerns, as customers can only guess what their data is used for.
Rethinking AI Solutions
As sensor technology keeps improving, their performance is increasing significantly, and their prices are decreasing, making them more accessible. However, the data generated by these sensors can quickly accumulate to several terabytes, making it difficult to transmit over a network, even with direct fiber optic cabling. This challenge suggests that data analysis is best done at the site of data generation, i.e., on the device.
In recent years, such local monitoring has become achievable through semiconductor technology advancements, which have enabled both the sensor and AI to be merged within one small circuit board. Therefore, this local data processing eliminates the need for data to be transmitted from sensor to server, while it merely needs to be extended to an AI on the same board. Here, the AI considers every snippet of data in the Random Access Memory (RAM), analyses them and then discards the sensor raw data. Therefore, only the analysis results are transmitted, which, in the simplest cases, can be illustrated by a lamp on the machine lighting up in the case of malfunction events. Similarly, the service can be notified directly that the machine or its components will experience anomalies or failures. The service can then identify the cause of the malfunction and schedule a maintenance appointment that does not disrupt production processes.
"Local AI models that are designed to work on a particular device or system are known as Embedded-AI systems," said Gromov. "Due to their inherent resource limitations, these systems are comparatively cost-effective at greater robustness. They also do not incur subsequent costs, such as those associated with network infrastructure, and are additionally capable of real-time operation, which can be crucial in safety-critical environments."
Potential Applications of Embedded AI
While the application possibilities of Embedded-AI are virtually boundless, the following comprises a selection of the technology use cases in various industrial scenarios:
- Monitoring motor drive shafts with hypersonic sensors, while the Embedded-AI can anticipate malfunction based on pattern deviation
- Safeguarding pumps and hoses with AI to identify material inconsistencies preemptively
- Utilizing pressure, vibration or acoustic sensors to determine the condition of axles and damper employing spectrographic sensors for the early detection of wear on conveyor belts
- Monitoring of main failure components and wear parts in machines
- Monitoring of cooling systems and heating elements.
Malfunction Prevention
In the field of maintenance services, predictive maintenance can improve reliability and quality while reducing costs by eliminating unnecessary service intervals. Manufacturers can thus offer greater reliability and quality with fewer staff and at lower cost.
This new trend simultaneously allows for further innovative possibilities besides a reduction of service intervals and guaranteeing failure safety. Embedded AI also opens up avenues for new business models, such as machine leasing instead of outright sale, aligning with manufacturers interests in long-term durability and sustainability. Customers would also greatly benefit from such models, as machine procurement does not have to be in the form of an expensive one-time investment, easing their liquidity.
"This approach is particularly advantageous in contemporary markets, where the demand for sustainable solutions and the challenge of skilled labor shortages are prevalent," said Gromov. "By adopting Embedded-AI, manufacturers can thus secure competitive advantages, ensure reliability and differentiate themselves from mass-market alternatives, such as those in Asia."
Authored by AITAD
For more information contact:
AITAD
www.aitad.de