Failure mode classification for condition-based maintenance in a bearing ring grinding machine

In a production environment, a CBM setup needs to be made available that enables predictive maintenance to be tried and developed. As discussed, the challenge not only lies in choosing the right methodology to analyze machine data but also in identifying the best way to monitor the machine and acquire data. The setup has to account for the choice of subsystems and machine components to be monitored and the intended data processing that will be required to develop the diagnostic and prognostic algorithms. The presence of cyber-infrastructure using information and communication technologies (ICT) is crucial in setting up a complete CBM system. Therefore, it has been demonstrated in our work [25] that a functioning and capable system is achievable when a production machine is equipped with sensors for condition monitoring in addition to the already existing process control setup. The CBM system has the physical elements and the required infrastructure to support the communication and interaction between different components. The data flow between different parts of the system is presented in Fig. 2.

The bearing ring grinder, top left in Fig. 2, is an external grinder SGB55 from Lidköping and is being used to grind outer rings of bearing type SKF-6210. The grinding machine has been equipped with sensors for process control and machine condition watchdogs. Table 1 lists standard sensors and standard systems incorporated in the process cycle recipe used to operate machines to grind bearing rings during their production. The sensors are mounted on the machine subsystems to measure their respective quantities before and during the grinding cycle. For example, the balancing unit is activated every time the grinding wheel is dressed to ensure a balanced grinding wheel and avoidance of vibrations causing quality issues. Similarly, acoustic emission, force, and power sensors are used to monitor and control the cycle itself. The machine’s control system is the SIEMENS Solution Line series Numerical Controller based system. The process data is captured using OPC server client communication and is published to monitor the machine’s state and output efficiency in the factory. The data is generated for every grinding cycle in the machine and is pushed into the database.

The additional sensors, listed in Table 2, are added to broaden the availability of data for the machine, process as well as condition monitoring. The machine and its subsystems, where the sensors are installed, are represented in the schematic view in Fig. 3. The sensor data is acquired in synchronization to the machine cycle using National Instruments’ (NI) data acquisition hardware where the control and acquisition software is developed in the NI LabView platform. The data acquisition system initiates sampling of sensor signals, at 100kHz, every time the machine goes into grinding mode and records the entire dataset from all the sensors for every produced part. The sensor data, along with the machine’s process data, is gathered from all the test runs before further processing.

A 2-stage position controlled grinding cycle is used with predefined cutting speed and feedrate [26]. The grinding cycle consists of 1. Rough and 2. Sparkout stages are shown in Fig. 4. A fixed approach known as the air grinding stage is used where the grinding slide approaches the ring with a higher speed than the grinding feedrate. This stage is programmed with the grinding allowance available to account for the incoming size and position of the ring, until close to contact, in the workhead tooling. Typical grinding force and power measured during the corresponding stages of the programmed grinding cycle are shown in Fig. 5. The herein described grinding cycle does not use in-process diameter measurement and in order to achieve a fixed output diameter, the machine is preconditioned and the cycle is adjusted to ensure that the desired dimension is produced for the incoming rings. This, despite using a simpler grinding cycle, is to ensure that the production machine setup procedure is followed and the rings are ground to the same dimensions as in production. Therefore, the quality of the produced rings in the experimentation can be measured using company standard measurement equipment.

For this study, the machine is prepared to produce parts for the selected cycle type under predefined machine conditions to replicate failures experienced in production. The failure modes are identified, after interviewing the maintenance technicians, from the maintenance history of similar machines in the factory. The top five frequently occurring faults, which have a direct impact on the produced quality and require maintenance intervention, are selected. Table 3 lists the sequence of experimental test runs for respective failure modes. Baseline tests in the beginning and at the end of the experiment ensure the reference machine operation is achieved against which the failure condition performance is evaluated.

Baseline tests are conducted by verifying the machine is in the standard operating condition by setting up the machine to produce parts in regular production. Belt Damage test is run after replacing the workhead drive belt with an older worn-out belt with enhanced damage as seen by maintenance technicians in production. Unbalance of the workhead spindle is achieved by adding unbalanced weight on the drive pulley. 2.5G of unbalance is introduced in the spindle according to ISO 1940-1 which is outside the specification of a grinding spindle. Setup - Drive plate is the run-out error in the direction of the ring face that can affect the ring rotation during the process. Setup - Workhead tooling fault is the error introduced in setting up the microcentric tool holder, used for ring chucking, as part of the machine resetting. This can happen due to parts miss-match or human error during the resetting process. Worn out - Workhead tooling support failure mode refers to the worn-out tool components in the microcentric tool holder. In this failure mode, damaged ring support (shoes) is installed that had been replaced earlier during the machine maintenance.

During normal production, the incoming un-ground rings have variations in terms of roundness and grinding allowance which can cause significant variations in the cycle behavior towards final output quality. Although this is accounted for by using a multi-stage grinding cycle, however, to keep consistency throughout the experimental test runs, the input rings are pre-rough ground to have an equal grinding allowance for all the test runs. Figure 6 shows the rings where a is the incoming hardened rings also known as black rings. After a little bit of stock removal, \(pre-rough\), b, and rings with the same diameter are achieved. These rings are used for the tests with the remaining grinding allowance of approx. \(500\mu m\) on diameter.

It is important to gather enough data to statistically balance out the variations within the tests and have equally distributed data to avoid over representation of any particular test type. Hence, for each test, a dressing interval of 15 rings is used where the grinding wheel is dressed after every \(15^{th}\) grinding cycle. In each failure mode test, a few dressing intervals are used to adjust the grinding cycle to produce rings in the desired dimension. This allowed the machine to be properly preconditioned and to reach a level where parts are consistently produced with the same amount of material removal. After reaching steady production, a series of 7 dressing intervals comprising 105 rings are produced. The total number of rings produced in this experimentation becomes 735 as depicted in Fig. 7. This procedure is repeated for each failure mode test and a set of rings thus produced are labeled for traceability and stored for measurement and future analysis. As described in Sect. 2.1.1 the data is simultaneously acquired for each grinding cycle and for individual rings produced in the machine. MATLAB is used as the main analysis tool for data exploration and processing as well as for analytical framework development.

Taking advantage of the knowledge of the cycle helps build a better predictive model [18]. Hence the grinding cycle for each ring is segmented into cycle sections based on the start of grinding before the grinding wheel makes the contact, force transient where the force in the system starts to rise just after the contact before reaching grinding steady state, grinding slide travel during rough grinding and finally sparkout at the end of the grinding cycle. Combining the change in gradient in time series signal of acoustic emission sensor with the process parameters from machine data, the segments are calculated for the individual ground rings. The contact detection at the beginning of the grinding cycle is used to calculate the actual length of the grinding cycle using true feedrate and the length that the grinding slide travels to grind the ring. The change of gradient in the signal data help identify the steady state of the grinding forces. Hence, these segments, for the typical force curve in the used grinding cycle, are presented in Fig. 8. This segmentation divides the time series data from each sensor for the individual ring into 4 segments. The change in the cycle of time-domain signal from the Acoustic Emission sensor is used to identify the segments for each ring. The building forces part also referred to as the force build-up segment, of the cycle is omitted due to the presence of transients resulting from the initial contact between the grinding wheel and the ring. The behavior of this transient segment is highly dependent on the incoming quality and any variation in the incoming quality will change the transient behavior. During production, it is expected to have variations in the pre-ground rings which can add unintended variations to the data and extracted features without giving any useful information. Thus the force build-up segment is not considered in feature extraction. The 3 segments of the cycle which are considered for further processing are listed in Table 4. From each individual sensor signal segment, time domain, as well as frequency domain features, is extracted.

A number of feature selection techniques like Fisher score [29] and chi-squared [30] can be used for the type of classification problem at hand. Also to keep the feature set representative of their physical source, principal component analysis is not considered a dimensional reduction step. Instead, MATLAB implementation of neighborhood component analysis (NCA) [31] is used for feature selection due to its computational efficiency and insensitivity towards irrelevant features in high dimensional feature space. The NCA algorithm in Matlab determines the feature weights by using a diagonal adaptation of NCA with regularization while minimizing an objective function that measures the average leave-one-out classification loss over training data. The output of the NCA is the weight vector that gives minimum classification error. The feature set is then sorted in descending order of NCA weights from where the top features are selected for sensor ranking.

Based on the feature ranking achieved using NCA, a sensor ranking criteria is defined as depicted in Algorithm 1 This results in sensors being weighted based on their features and frequency of occurrence in the top feature list. The sensor ranking is achieved through sorting the weights. The sensors can be chosen by taking advantage of the domain knowledge in setting up the grinding process in the machines. Thus in addition to ranking, the selection process can also incorporate the cost and complexity of sensors installation and their relevance to the process and targeted fault diagnosis. The dropping of sensors reduces the feature set further to improve the selection of simpler classification models with reduced complexity and better overall performance through increased generalization without having to train on extremely large datasets. Since there can be various factors deciding on sensor selection thus the final list cannot be deterministic. The final sensors list for the above mentioned criteria is presented later in Table 10 in Sect. 3.

For failure diagnostics in CBM, supervised learning models like support vector machines (SVM) [20], k-nearest neighbor [32], neural networks [33], and decision trees [9] can be used for classification. However, for the scope of this work, the classifiers in Matlab’s classifier app are benchmarked using the training dataset for the multi-class classifier. Therefore, the top performing random forest using the default hyper-parameters as listed in Table 6 is selected as the prediction model for both binary as well as multi-class classifiers. The tuning of these hyper-parameters is not considered in this work.