One major challenge for any enterprise with production-critical and expensive machines is to avoid their very costly, catastrophic failures or breakdowns. In this regard, the reactive maintenance of machinery that has already failed is not appropriate, and the preventive replacement of expensive machine parts at set time intervals is often neither affordable nor effective for avoiding unknown or unpredictable failure modes. The only alternative maintenance strategy that solves the problem is condition-based maintenance with which machinery will be maintained only when needed with regard to their actual condition. This requires the detection of the onset of machine failures before they occur, which is the ultimate goal of monitoring the condition of machines.
Condition monitoring (CM) can be defined as the field of technical activity in which selected physical parameters associated with machinery operation are periodically or continuously observed for the purpose of determining the machinery integrity. As such it is the prerequisite of condition-based maintenance: The assessment of the actual condition of the operating machine allows for the early detection and diagnosis of developing faults which can then be corrected with respectively informed and timely scheduled maintenance activities. In this respect, condition monitoring can help to improve the safety of the machine as well as to avert unnecessary downtimes and a decrease of product quality caused by machine faults. The installation of an appropriate machine-specific condition monitoring system (CMS) in combination with on-condition remote or mobile on-site maintenance services can therefore lead to a significant reduction of maintenance and production costs. Such systems are commonly in use in many enterprises of manufacturing and processing industries such as the steel and paper production, automotive, aerospace and wind energy industry.
The process of condition monitoring encompasses the steps of data collection, fault detection and diagnosis where the first two correspond with machine inspection. First, condition-relevant data is collected by non-destructively measuring selected operation parameters like temperature, pressure, vibration, or lubrication with appropriate instruments and transducers (sensors). The presence of faults, that is, a faulty condition of the monitored machinery, can then be detected by evaluating these data against respective, machine-specific acceptance limits and standards of the parameter values for different degrees of machine conditions. A subsequent fault diagnosis is concerned with the recognition of the type and location of a detected fault. However, condition monitoring is sometimes identified with mere inspection.
CM technologies. Principal technologies for condition monitoring are vibration analysis, infrared thermography, oil and wear debris analysis, acoustic monitoring, and visual inspection.
Acoustics-based condition monitoring of machines bases on the measurement of noise intensities or sound vibration patterns with techniques similar to vibration analysis.
Visual inspection of machines is the most rudimentary form of condition monitoring and typically carried out by experienced human technicians on-site with all human senses (hear, see, smell, taste, touch) and supported by portable instruments that offer some of the aforementioned technologies.
Vibration analysis. Faults and damages of a machine like unbalance, misalignment, looseness or structural resonances will cause it to vibrate with a certain frequency and strength which can be non-destructively measured with accelerometers and sensors attached to the machine. These vibration signals are represented in the time and the frequency domain. In the first case, the time waveform of machine vibration can be used to detect a critical or faulty condition by checking whether the maximum amplitude does exceed the machine-specific, standard limit for such a condition. The condition trend on a larger time scale can be displayed with a compressed time waveform that is the time series of indices of smoothed time waveforms of which it is composed, where extrapolation is used to estimate when the warning level will be reached which is the optimal point for on-condition maintenance. For the purpose of fault diagnosis, the analog vibration signals are typically processed into the frequency domain by means of a discrete and fast Fourier transform. As a result, superposed vibration signals of a composite time waveform are separated according to their individual frequencies in the signature spectrum of the machine. The type of the fault can be determined by comparing the actual signature spectrum with machine-specific reference spectra of fault types.
Infrared thermography determines the surface temperatures of a machine by measuring the intensities of its emitted infrared energy, displays them in a thermal image and compares the values with given limits to detect a faulty condition. The colored display of temperature regions supports visual fault diagnosis. Alternatively, oil and wear debris analysis determines the wearing condition of the machine based on the quality of an appropriate lube oil sample in terms of its viscosity, oxydation, acidity, and physical or chemical contamination. For example, the number, shape, size and chemical composition of metal particles in a contaminated sample indicate from where they have been rubbed or sliced off, thus the fault type and its possible location.
There is no general-purpose rule for CM technology selection. Although vibration analysis is covering many prominent fault types not only of rotary machines, it is widely recommended to incorporate as many other CM technologies as possible into the inspection and diagnosis to assure reliable results. For example, the worn out condition of a roller bearing can be detected based on its abnormal vibration and confirmed by detecting its lube oil contamination and overheating. The economic trade-off of CMS is another selection factor.
One research trend is intelligent condition monitoring by utilizing methods from computer science, especially AI. For example, automated fault type classification can be achieved either multi-parametric data-driven with neural networks and support vector machines or domain knowledge-driven with ontology reasoners. Intelligent self-monitoring and control of machines can be implemented with intelligent agents, wireless M2M communication and semantic services. Interactive web-based 3D simulation of machine conditions and planned maintenance steps is of added value for teleservice partners and realizable with 3D web, semantic web and AI technologies
CREMA investigates means of (vertically) integrating means of intelligent condition monitoring (iCM) with process optimisation and management in the cloud. That includes: means for multi-variate sensor data stream processing in the cloud for fault detection and diagnosis online, the combination of statistical, semantic and probabilistic analysis with semantic explanation to the user for this purpose, and the integration of iCM with dynamic optimisation of affected manufacturing processes in the cloud.