Artificial intelligence has crept into almost every aspect of modern society. The recent boom in AI started with a subcategory called Machine Learning (ML), where software is ‘taught’ how to recognize patterns in data using sophisticated algorithms that can predict future outcomes on new data sets. Although pattern recognition is nothing new from an application perspective, the advantages of ML over traditional coding is that the algorithm’s prediction accuracy continues to increase as it ‘sees’ more data – without needing to be explicitly told to do so.
The strengths of ML apply perfectly to predictive maintenance (PdM), possibly its first real success story, where other emerging technologies such as IoT and cloud data storage merge with ML models to create a pipeline of data collection, storage, and processing. This new methodology lends itself to manufacturing, where equipment stability and accuracy are paramount to the success of a business.
PdM is a method of using data to anticipate maintenance requirements on the shop floor before a failure mode occurs. PdM relies on specific data being pulled from the equipment in near real time for detection. This can include a variety of information, from temperate sensors in electrical cabinets or coolant tanks, to vibration sensors on spindles or tool stations, to direct load monitoring of axis output from the CNC controller itself. The ML model can act as a safe-guard for live processes – notifying production managers, maintenance personnel, and technicians when a specific machine or component is going to fail, and also be used to make existing quality and production systems more robust by retroactively analyzing data to tune manufacturing parameters.
Some of the many benefits to implementing a PdM system are outlined below along with a high-level look at investment and scalability options.
PdM can optimize machine downtime by proactively addressing machine issues before they lead to process failure. Traditionally planned downtime is limited to preventative maintenance, but just like greasing a machine or topping off an oil reservoir, heavy maintenance items like ballscrew replacement or time-consuming activities like axis referencing and tramming, can be scheduled when they have the least impact on production orders.
With optimized downtime comes the ability to negate unplanned downtime, allowing for longer, more consistent runs of a production line and less fire-fighting from supervisors scrambling to get replacement parts and reallocate production personnel
Depending on the nature of the telemetry data being collected, PdM can also act as a process control system. Data from a machining cycle can be used to help determine tool life management or ideal cutting conditions.
PdM is also an investment for future business opportunities. As equipment ages the machine responds differently to production stresses and the associated maintenance schedule must be updated. as. Rather than making reactive decisions, data exhibiting an increased frequency of poor cutting conditions and rising repair costs are used to help upper management predetermine when a machine tool is at the end of its useful life, allowing engineering and shop supervisors time to decide the best course of action.
Improving product quality is in every company’s quality statement. A machine tool kept in top condition, outputs a higher quality product over its lifetime. Using PDM to prevent failures that damage a machine tools integrity, keeps that tool in better condition as compared to one resigned to preventative maintenance.
For companies willing to invest in PdM there are options to get started from high end – complete package solutions, to custom or ‘home-grown’ installations.
An example of a complete solution would be ADLINK’s smart gateway. ADLINK teamed up with Intel, IBM, and PrismTech and developed a product that gathers data from sensors in the factory and performs what is known as edge analytics. ADLINK uses Intel’s standard architecture and security features for consistency and performance. PrismTech supplies the means for data distribution making it easy to connect with existing company networks and add new devices. And lastly IBM’s PMQ software provides the predicative model for analysis.
A mid-tier solution may be using software that’s been available for years from controller manufacturers, like Fanuc, to output extensive data on a machine tools condition. Fanuc’s MTLINK i option is capable of dumping axis data like position, load, temperature, and overtravel occurrences as well as spindle data like RPM, load, and override settings. This type of solution requires a skilled analyst to collect the data and develop ML models for predictive maintenance.
Last is the ‘home-grown’ option which is by far the cheapest but requires all the development work to be performed in house. This is a viable solution for shops that have more antiquated machinery. A Raspberry Pi and Arduino can act as a sensor node and data gateway, taking frequent sensor readings and appending this information to a cloud database. I performed an exercise to this end where sensors were read every three seconds, once an hour this data was appended to a database in my DropBox automatically through their API. From there the data just needs to be processed by a ML model and a course of action decided for notification purposes to close the loop.
There are many approaches to PdM making it possible to find a solution that best fits your environment. AI will continue to grow rapidly, making solutions more accessible and more adaptable. Although it may it seem intimidating, and conjure up images of robots running your factory, the opportunity to learn and incorporate AI into your business, is fun and rewarding. It’s in your companies’ interest to embrace these developments and quickly learn how to reap the benefits of the unavoidable growth in AI.
Chris Macaluso, February 2019
CNC Programmer
Artificial intelligence has crept into almost every aspect of modern society. The recent boom in AI started with a subcategory called Machine Learning (ML), where software is ‘taught’ how to recognize patterns in data using sophisticated algorithms that can predict future outcomes on new data sets. Although pattern recognition is nothing new from an application perspective, the advantages of ML over traditional coding is that the algorithm’s prediction accuracy continues to increase as it ‘sees’ more data – without needing to be explicitly told to do so.
The strengths of ML apply perfectly to predictive maintenance (PdM), possibly its first real success story, where other emerging technologies such as IoT and cloud data storage merge with ML models to create a pipeline of data collection, storage, and processing. This new methodology lends itself to manufacturing, where equipment stability and accuracy are paramount to the success of a business.
PdM is a method of using data to anticipate maintenance requirements on the shop floor before a failure mode occurs. PdM relies on specific data being pulled from the equipment in near real time for detection. This can include a variety of information, from temperate sensors in electrical cabinets or coolant tanks, to vibration sensors on spindles or tool stations, to direct load monitoring of axis output from the CNC controller itself. The ML model can act as a safe-guard for live processes – notifying production managers, maintenance personnel, and technicians when a specific machine or component is going to fail, and also be used to make existing quality and production systems more robust by retroactively analyzing data to tune manufacturing parameters.
Some of the many benefits to implementing a PdM system are outlined below along with a high-level look at investment and scalability options.
PdM can optimize machine downtime by proactively addressing machine issues before they lead to process failure. Traditionally planned downtime is limited to preventative maintenance, but just like greasing a machine or topping off an oil reservoir, heavy maintenance items like ballscrew replacement or time-consuming activities like axis referencing and tramming, can be scheduled when they have the least impact on production orders.
With optimized downtime comes the ability to negate unplanned downtime, allowing for longer, more consistent runs of a production line and less fire-fighting from supervisors scrambling to get replacement parts and reallocate production personnel
Depending on the nature of the telemetry data being collected, PdM can also act as a process control system. Data from a machining cycle can be used to help determine tool life management or ideal cutting conditions.
PdM is also an investment for future business opportunities. As equipment ages the machine responds differently to production stresses and the associated maintenance schedule must be updated. as. Rather than making reactive decisions, data exhibiting an increased frequency of poor cutting conditions and rising repair costs are used to help upper management predetermine when a machine tool is at the end of its useful life, allowing engineering and shop supervisors time to decide the best course of action.
Improving product quality is in every company’s quality statement. A machine tool kept in top condition, outputs a higher quality product over its lifetime. Using PDM to prevent failures that damage a machine tools integrity, keeps that tool in better condition as compared to one resigned to preventative maintenance.
For companies willing to invest in PdM there are options to get started from high end – complete package solutions, to custom or ‘home-grown’ installations.
An example of a complete solution would be ADLINK’s smart gateway. ADLINK teamed up with Intel, IBM, and PrismTech and developed a product that gathers data from sensors in the factory and performs what is known as edge analytics. ADLINK uses Intel’s standard architecture and security features for consistency and performance. PrismTech supplies the means for data distribution making it easy to connect with existing company networks and add new devices. And lastly IBM’s PMQ software provides the predicative model for analysis.
A mid-tier solution may be using software that’s been available for years from controller manufacturers, like Fanuc, to output extensive data on a machine tools condition. Fanuc’s MTLINK i option is capable of dumping axis data like position, load, temperature, and overtravel occurrences as well as spindle data like RPM, load, and override settings. This type of solution requires a skilled analyst to collect the data and develop ML models for predictive maintenance.
Last is the ‘home-grown’ option which is by far the cheapest but requires all the development work to be performed in house. This is a viable solution for shops that have more antiquated machinery. A Raspberry Pi and Arduino can act as a sensor node and data gateway, taking frequent sensor readings and appending this information to a cloud database. I performed an exercise to this end where sensors were read every three seconds, once an hour this data was appended to a database in my DropBox automatically through their API. From there the data just needs to be processed by a ML model and a course of action decided for notification purposes to close the loop.
There are many approaches to PdM making it possible to find a solution that best fits your environment. AI will continue to grow rapidly, making solutions more accessible and more adaptable. Although it may it seem intimidating, and conjure up images of robots running your factory, the opportunity to learn and incorporate AI into your business, is fun and rewarding. It’s in your companies’ interest to embrace these developments and quickly learn how to reap the benefits of the unavoidable growth in AI.
Chris Macaluso, February 2019
CNC Programmer
Artificial intelligence has crept into almost every aspect of modern society. The recent boom in AI started with a subcategory called Machine Learning (ML), where software is ‘taught’ how to recognize patterns in data using sophisticated algorithms that can predict future outcomes on new data sets. Although pattern recognition is nothing new from an application perspective, the advantages of ML over traditional coding is that the algorithm’s prediction accuracy continues to increase as it ‘sees’ more data – without needing to be explicitly told to do so.
The strengths of ML apply perfectly to predictive maintenance (PdM), possibly its first real success story, where other emerging technologies such as IoT and cloud data storage merge with ML models to create a pipeline of data collection, storage, and processing. This new methodology lends itself to manufacturing, where equipment stability and accuracy are paramount to the success of a business.
PdM is a method of using data to anticipate maintenance requirements on the shop floor before a failure mode occurs. PdM relies on specific data being pulled from the equipment in near real time for detection. This can include a variety of information, from temperate sensors in electrical cabinets or coolant tanks, to vibration sensors on spindles or tool stations, to direct load monitoring of axis output from the CNC controller itself. The ML model can act as a safe-guard for live processes – notifying production managers, maintenance personnel, and technicians when a specific machine or component is going to fail, and also be used to make existing quality and production systems more robust by retroactively analyzing data to tune manufacturing parameters.
Some of the many benefits to implementing a PdM system are outlined below along with a high-level look at investment and scalability options.
PdM can optimize machine downtime by proactively addressing machine issues before they lead to process failure. Traditionally planned downtime is limited to preventative maintenance, but just like greasing a machine or topping off an oil reservoir, heavy maintenance items like ballscrew replacement or time-consuming activities like axis referencing and tramming, can be scheduled when they have the least impact on production orders.
With optimized downtime comes the ability to negate unplanned downtime, allowing for longer, more consistent runs of a production line and less fire-fighting from supervisors scrambling to get replacement parts and reallocate production personnel
Depending on the nature of the telemetry data being collected, PdM can also act as a process control system. Data from a machining cycle can be used to help determine tool life management or ideal cutting conditions.
PdM is also an investment for future business opportunities. As equipment ages the machine responds differently to production stresses and the associated maintenance schedule must be updated. as. Rather than making reactive decisions, data exhibiting an increased frequency of poor cutting conditions and rising repair costs are used to help upper management predetermine when a machine tool is at the end of its useful life, allowing engineering and shop supervisors time to decide the best course of action.
Improving product quality is in every company’s quality statement. A machine tool kept in top condition, outputs a higher quality product over its lifetime. Using PDM to prevent failures that damage a machine tools integrity, keeps that tool in better condition as compared to one resigned to preventative maintenance.
For companies willing to invest in PdM there are options to get started from high end – complete package solutions, to custom or ‘home-grown’ installations.
An example of a complete solution would be ADLINK’s smart gateway. ADLINK teamed up with Intel, IBM, and PrismTech and developed a product that gathers data from sensors in the factory and performs what is known as edge analytics. ADLINK uses Intel’s standard architecture and security features for consistency and performance. PrismTech supplies the means for data distribution making it easy to connect with existing company networks and add new devices. And lastly IBM’s PMQ software provides the predicative model for analysis.
A mid-tier solution may be using software that’s been available for years from controller manufacturers, like Fanuc, to output extensive data on a machine tools condition. Fanuc’s MTLINK i option is capable of dumping axis data like position, load, temperature, and overtravel occurrences as well as spindle data like RPM, load, and override settings. This type of solution requires a skilled analyst to collect the data and develop ML models for predictive maintenance.
Last is the ‘home-grown’ option which is by far the cheapest but requires all the development work to be performed in house. This is a viable solution for shops that have more antiquated machinery. A Raspberry Pi and Arduino can act as a sensor node and data gateway, taking frequent sensor readings and appending this information to a cloud database. I performed an exercise to this end where sensors were read every three seconds, once an hour this data was appended to a database in my DropBox automatically through their API. From there the data just needs to be processed by a ML model and a course of action decided for notification purposes to close the loop.
There are many approaches to PdM making it possible to find a solution that best fits your environment. AI will continue to grow rapidly, making solutions more accessible and more adaptable. Although it may it seem intimidating, and conjure up images of robots running your factory, the opportunity to learn and incorporate AI into your business, is fun and rewarding. It’s in your companies’ interest to embrace these developments and quickly learn how to reap the benefits of the unavoidable growth in AI.
Chris Macaluso, February 2019
CNC Programmer