If there is one thing we can say for sure about machines it is that sooner or later they will fail and cause problems. Machines have mechanical, electrical and electronic systems that wear out and can fail with time and use. These failures can be aggravated by various reasons, such as: operator error (human factor), improper use, poor quality consumables (belts, drills, lubricant, etc.) or duty cycle outside the manufacturer’s specifications. The maintenance of the machines is crucial so that they do not stop beyond what is necessary and production remains normal. The abnt NBR 5462: 1994 standard, which deals with reliability and maintainability, defines three types of maintenance:
- Corrective maintenance: the most common in the industry, occurs after failure, reacting to occurrence. In this case, it is necessary to arrange the repair of the machine in the shortest possible time, so that the production returns to normal;
- Preventive maintenance: done at scheduled intervals, seeking to reduce the probability of failures with the verification of the operation of the machines, measurement and correction of deviations and exchange of items with wear and prone to failures. The goal with this type of maintenance is to reduce the need for future corrective maintenance;
- Predictive maintenance: the most modern approach, in which data are collected on the operation of machinery during its normal use. Deviations in the collected data may indicate an imminent or future failure. The goal is to reduce preventive maintenance to a minimum and decrease corrective maintenance.
In predictive maintenance, the new concept of machine monitoring arises, which means providing machine sensing to generate data and apply tools to the analysis of these data, to predict failures or improve their performance. The definition of which sensors will be used and which analyses will be made depends on the characteristics of the machine to be monitored and the production process in which it is used.
There is a learning curve for collection and analysis to have an effect on reducing the need for preventive and corrective maintenance. When an employee is hired to operate a machine that performs a certain process in the industry, it takes some time for him to adapt to the process and operation of the equipment. Over time, he learns the behavior of the machine and can, based on different noises or vibrations at some points of the equipment, point out the problems in the machinery before they occur.
Machine monitoring works much like the newly hired employee. First, we deploy sensors at predetermined points for data collection. We send this data to be recorded with the machine running normally for some time and use this data for future comparative analysis (learning). After some time, we are able to know, from the collected data, the normal behavior of the machine.
Then, just as the employee hearing a different noise coming from the machine perceives something strange, the comparative analysis of the monitoring indicates that something wrong — a failure — is about to happen.
Because the sensors are connected directly to the machines, monitoring important points throughout the process, more data is collected, forming a more complete and detailed history of the operation of the equipment. With them, the preventive maintenance team is able to schedule checks more accurately.
How machine monitoring works
The types of sensors that will be employed depend on the equipment to be monitored. Since electric motors are present in virtually all machines, the most common is the use of sensors in the power supply lines, measuring the voltage and electric current consumed.
By analyzing the current consumption of the motor, we can detect noises, peaks or imbalances in the phases. This data may indicate that something is wrong on the machine. A mechanical overload in the engine, for example, causes the appearance of variations in power consumption, which in turn can be seen and analyzed to indicate that the equipment needs maintenance.
These types of signals can also start small and grow over time, something that can be accompanied by monitoring and signaled to a technician when the variations exceed a previously determined value. In this type of situation, the team has both the data that caused the alarm and the historical progressive growth of the equipment signals, which can also be analyzed to indicate the operation of the machines.
More recently, it has also become common to use accelerometers for vibration monitoring (link 1). An accelerometer measures the variations, even if very small, in the acceleration of a specific point on the three axes of Space (X, Y and Z). Any movement, in any direction, generates a variation in the values read by the sensor. In normal operation, the acceleration behavior is known and recorded. In this way, deviations can be detected and analyzed more easily.
Some more specific industrial processes may make use of other sensors, such as temperature sensors in plastic injection machines. Temperature control in the molds is a fundamental process to avoid failures, as possible future failures can be predicted if the temperature curve in the molds undergoes some change over time.
More modern machines usually already have the integrated sensors and deliver the data via an Ethernet port, USB or other format. But any machine can be monitored, provided that a study is made of the best points and types sensors that will be employed. In this case, it is also necessary to install an electronic system for collecting the data from the sensors and sending the data for analysis. With the miniaturization of the sensors, even smaller power tools can be monitored, such as hand drills and screwdrivers.
Once the data is captured, it should be sent for external analysis. The simplest way is to put this data on a screen (dashboard) for viewing. In this case, a member of the maintenance team can monitor and check deviations in the values read through graphs or tables, comparing it to the previous history of machine readings.
This function can also be automated, with software doing the analysis — from simple comparison to normal machine operation data to analysis with technologies such as machine learning and artificial intelligence to predict failure trends. Since the data is collected for a long time, a good database is created (dataset) with the normal behavior of the machine. Each detected deviation can indicate a fault, which must be checked with preventive maintenance and then catalogued and detailed in the database-creating a positive feedback in the process.
With time and learning the algorithm, the deviations will be interpreted by the monitoring system itself and it will indicate what the possible failure, how to solve and what material will be needed for maintenance.
Machine monitoring has come to complement and increase the productivity of maintenance teams, who now rely on means to analyze and predict failures before they happen. Staff can better schedule preventive maintenance and decrease corrective maintenance.
Automatic data analysis, done by artificial intelligence and machine learning, creates a history of possible machine abnormalities, their probable cause, and directives for correction. Maintenance managers can perform their work faster, as they already know in advance what they need to change or adjust so that the machine returns to normal operation.
In addition to failure prediction, machine monitoring also allows you to discover opportunities for improvement. Using the history of the monitored behavior of the machinery, it is possible to perform tests with consumables (lubricants, for example) of different makes and models and, through comparison, to verify which one has the best performance.