Home > Article > Technology peripherals > "Focus on mature machine learning applications with investment value"
The Internet of Things (IoT) has grown steadily for more than a decade, driven by the proliferation of connected devices. Today's billions of connected devices provide businesses with unprecedented opportunities to collect and analyze data from the physical world to improve their business processes. In some cases, they are also driving novel and successful business models, leading enterprises to ride the wave of IoT adoption.
In most cases, enterprises process IoT data in edge clusters or clouds, not in edge devices and microcontrollers. The emergence of embedded machine learning and TinyML disrupts this paradigm, pushing application intelligence to the edge of the IoT network. As mentioned in the first article in this series, this provides significant benefits, including:
The benefits are tangible and have clear business relevance . The ability for enterprises to use more data and processes at the edge of the network can increase business efficiency, which directly translates into monetary gains and improved corporate profits. Embedded machine learning is a game changer in artificial intelligence and IoT computing that can increase enterprise productivity. Here are 5 use cases for industrial enterprises deploying embedded machine learning.
Most industrial enterprises maintain assets based on a preventive maintenance approach, which depends on regular maintenance or replacement of machines and tools, etc. These intervals are determined by the maintenance policy provided by the equipment manufacturer. This approach helps avoid catastrophic production downtime events because assets are often maintained before failure occurs. However, preventive maintenance results in less than optimal asset utilization because assets are always replaced prematurely.
The emergence of Industry 4.0 and the Industrial Internet of Things enables industrial companies to implement condition-based monitoring of their assets. Leveraging digital data from sensors (e.g. vibration sensors, temperature sensors, thermal images) and asset management systems, businesses can now gain real-time visibility into the status of industrial assets such as tools and machinery. Additionally, using machine learning algorithms, they can gain predictive insights into the remaining useful life (RUL) of their assets. In some cases, reliable RUL estimates enable industrial companies to transform preventive maintenance into predictive maintenance. Predictive maintenance is the ultimate vision for maintenance and repair operations to achieve optimal Overall Equipment Effectiveness (OEE). Condition-based monitoring and predictive maintenance can help enterprises improve asset utilization, reduce production downtime, eliminate waste caused by equipment failure, and schedule maintenance tasks at the optimal time. Predictive maintenance is considered one of the killer applications of the fourth industrial revolution (Industry 4.0): it has a tangible return on investment and is applicable to almost all industrial sectors, including manufacturing, energy, construction, smart buildings, oil and gas, and mining, etc.
Most predictive maintenance deployments transmit and analyze data in the cloud. This approach has operational limitations, for example, failure predictions from cloud-based machine learning analytics are not always fast enough to take appropriate remedial or preventive actions. Embedded machine learning adds significant value to predictive maintenance and condition monitoring: it generates real-time insights and enables real-time decision-making. Performing machine learning directly on a data acquisition device or microcontroller inside the machine enables industrial companies to gain timely and accurate visibility into the status of various assets. This unlocks the potential for real-time decision-making based on actual equipment conditions. Overall, embedded machine learning improves the efficiency of predictive maintenance applications, increases asset utilization, and optimizes the quality of their services.
Machine learning has recently opened new horizons for quality management in manufacturing and production operations. Specifically, it gives the concept of predictive quality, that is, the ability to predict quality problems before they occur. In this regard, machine learning techniques, including deep learning, are applied to the production line. The purpose of its algorithms is to proactively identify conditions or patterns that lead to product defects. Based on this, the plant manager can take remedial measures to prevent defects from occurring. Additionally, machine learning techniques can be used to optimize patterns related to other parameters such as cost and environmental performance.
Embedded machine learning adds significant value to the quality management use cases mentioned above. Specifically, it provides a way to extract predictive insights into potential defects based on the processing of data inside a device. These insights can be combined with information from cloud analytics to identify process and control parameters that lead to quality issues. Likewise, they can be used to optimize multiple parameters simultaneously, enabling zero-defect manufacturing. Embedded machine learning thus provides factory managers and quality engineers with real-time asset-level information about defects, which complements existing knowledge about quality management issues. Therefore, it enables businesses to excel in implementing quality management strategies such as Total Quality Management (TQM) and Six Sigma. Overall, industrial companies can leverage embedded machine learning to supplement their existing quality management knowledge to improve product quality while reducing production time and costs.
In recent years, the Internet of Things has had a transformative impact on smart building and facility management applications. Deploying sensors in buildings and other real estate assets allows property owners to access real-time, up-to-date information about the status of their properties. Based on this information, they can optimize the operation of HVAC (heating, ventilation and air conditioning) systems to save costs and improve their environmental indicators. In this regard, occupancy monitoring applications are very important.
Based on the processing of data from temperature and other sensors, an accurate understanding of the occupancy of rooms and other physical assets such as desks, computers and office spaces can be obtained. This is key to optimizing energy efficiency and maximizing tenant comfort. Additionally, it provides facility managers with real-time insights on asset utilization, allowing them to plan their usage and improve their overall productivity. Over the past few months, demand for such occupancy monitoring applications has surged due to the COVID-19 outbreak. The latter has led to the implementation of large-scale teleworking policies, which has made it more challenging for facilities managers to monitor and predict asset occupancy patterns. Sensors and IoT applications can help them by providing reliable and timely information about tenants’ physical presence in various spaces.
In facilities management environments, embedded machine learning improves the sustainability and accuracy of occupancy management applications. Specifically, it can run statistical data analysis within occupancy monitoring sensors without having to aggregate multiple sensor values through a cloud gateway. This improves the accuracy and timeliness of monitoring while also helping to reduce CO2 emissions. Embedded machine learning is important as facility managers are turning to IoT to reduce emissions and meet ambitious sustainability goals. In this way, they enhance their brand image and increase compliance with relevant regulations. For example, the recent New York City Climate Mobilization Act (CMA) requires buildings to be more energy efficient. Specifically, it mandates that buildings larger than 25,000 square feet must reduce greenhouse gas emissions by 40% by 2030 and by 80% by 2050, compared to 2005 levels. Overall, embedded machine learning is a powerful tool for next-generation energy-efficient facility management applications.
In the past few years, IoT systems and embedded devices have penetrated into the agricultural field and enabled precision farming. A prominent example is that sensors and ubiquitous connected devices such as beacons, RFID tags and specialized embedded sensors (such as stomach sensors) are increasingly being implanted in livestock to allow farmers to monitor them. To this end, relevant IoT applications tend to transmit raw data about the condition of cattle to the cloud for proper analysis. However, in some cases this approach may be inefficient or even unfeasible, as most cattle herds live in outdoor environments covering thousands of hectares in size. In this setting, network connectivity (e.g., short-range IoT networks) may not be sufficient to support the required quality of service during the data aggregation process. Furthermore, such devices often require battery power, which creates energy autonomy issues.
Embedded machine learning and TinyML provide substantial help in mitigating these limitations. Data analysis occurs on the livestock, which significantly reduces the amount of data that needs to be transferred to the application backend. Rather than collecting a continuous stream of data, deploying machine learning on embedded devices can stream data on a regular basis (e.g., every hour). This can provide farmers with insights into the condition of their animals and their activities (e.g., resting, suffering, or roaring). These insights enable farmers to make informed decisions about production processes such as milking and slaughtering. Overall, embedded machine learning helps farmers take advantage of precision livestock monitoring systems in situations where traditional cloud processing is impossible or ineffective.
Machine learning and computational intelligence techniques are also used in crisis management and civil defense applications, including earthquake and wildfire predictions. In this regard, data from various sensors are often aggregated and processed in the cloud. However, in crisis management, time is of the essence: the success of crisis management actions largely depends on the timeliness of crisis management indicator predictions. For example, identifying earthquake warning signs earlier could lead to faster, more effective action. This is an area where embedded machine learning is of great value.
When it comes to wildfire management, embedded machine learning can provide reliability and deployment advantages, similar to cattle monitoring situations. In particular, executing statistical models within embedded sensors could facilitate timely predictions of wildfires without the need for strong network connections and battery-powered devices.
Embedded machine learning is widely used, and its application scope is not limited to the above five. For example, in precision agriculture, it can detect crop diseases directly on crops without the need for Various data streams are aggregated and analyzed in the cloud. Another example is that it enables precise refrigeration intelligence applications that directly analyze the temperature of sensitive products such as food, beverages and pharmaceuticals without having to use ambient temperature to estimate temperature anomalies. Overall, embedded machine learning unlocks nearly unlimited opportunities for innovation in many different areas.
However, developing and deploying embedded machine learning applications in industrial environments is not easy. Each implementation step must be carefully planned to meet stringent industrial requirements. From choosing the right embedded device to obtaining enough training data and implementing the right machine learning model, developers and deployers must make careful choices.
The above is the detailed content of "Focus on mature machine learning applications with investment value". For more information, please follow other related articles on the PHP Chinese website!