Real-Time Envelope Monitoring of High-Speed Spindle in Commissioning Conditions: Grinding Machine

Abstract

This article addresses the monitoring and diagnosis of high-speed spindles (HSM) used in CNC grinding machines, emphasizing the importance of the real-time evaluation of their dynamic behavior during commissioning. Due to the complexity of these dynamic phenomena, especially at high speeds (up to 150,000 RPM), common faults such as bearing wear, imbalance, or misalignment can lead to catastrophic failures and high repair costs. An original method is proposed, based on synchronous envelope vibration analysis (SEVA) using the Hilbert transform, to detect mechanical defects in both low-frequency domains (imbalance, mechanical looseness) and high-frequency domains (bearing faults). The system includes vibration, temperature, and speed sensors, and the experimental protocol involves step-by-step monitoring from 10,000 to 90,000 RPM. Through synchronous FFT analysis and IFFT, critical frequencies and their impacts on machining quality are identified. The method enables the accurate fault diagnosis of new or refurbished spindles under real industrial conditions, reducing downtime and production losses. The method involves both local and remote real-time monitoring and diagnosis using a remote data center protocol.

1. Introduction

Scientific research in the field of CNC machine tools has registered high development in the last decade, moving from automatic to digital processes and control [1,2,3,4,5,6,7,8,9]. At the same time, the complexity of the dynamic phenomena encountered in the spindles of machine tools is well known, especially in the presence of high speeds [3]. In-depth knowledge of the dynamic phenomena generated by the cutting process and their influence on the behavior and lifespan of the high-speed spindle has become a key factor, especially with the digitalization of the mechanical processing industry [7]. With the increase in productivity and the reduction in downtime, the performance and reliability of the spindle are subject to a long development process [9]. The multitude of factors and their complexity can directly influence the quality of the spindle’s operation and, implicitly, the quality of the cutting process and the part [1,2,3].

The condition of the spindle determines the quality of machining and the performance of machine tools, but also the lifespan of certain machine tools. In general, the spindle is the element that provides the speed in cutting, being designed and built in different configurations for different types of machine tools depending on the precision, type of machine tool (milling, grinding, turning, drilling), conventional speeds or high speeds, etc. [3]. This paper focuses on high-speed spindles in integrated motor–spindle construction.

Taking into account the technical–functional and economic importance of the spindle—specifically the motor spindle—the development of models based on artificial intelligence in the scope of monitoring, diagnosis, and adaptive control in digital configurations in real time represents a crucial need [7,10,11,12,13]. However, in industrial conditions, models need time to reach optimal maturity; thus, it is necessary to acquire many datasets and simulations. In the case of commissioning, when running-in is needed, the models must have high accuracy, both qualitatively and quantitatively; the speeds are very high; and the decision must be made quickly and based on concrete results.

With the increase in productivity and the reduction in downtime, the performance and reliability of the spindle are subject to a long development process. The multitude of factors and their complexity can directly influence the quality of the spindle’s operation and, implicitly, the quality of the cutting process and the workpiece [14,15,16]. It is imperative to monitor the spindle condition, including the evolution of wear and bearing defects, before the spindle catastrophically fails, leading to major damage to component elements and the machine, as well as production losses [17,18]. Many works are available in the literature about monitoring and fault diagnosis in machine tools, including for the spindle [19,20]. However, in general, they only deal with a single problem or with certain problems in laboratory conditions, on test benches, or in specific test installations [4,6,8,10,21]. Instead, in real cases, multiple faults are quite common: the simultaneous presence of a bearing fault associated with an imbalance or a misalignment, an electrical fault, or a worn gear frequently occurs in the spindles of machine tools. The effective monitoring and accurate diagnosis of spindle wear are essential. Monitoring means providing timely warnings when a defect occurs, and diagnosis means correctly identifying the type of defect for intelligent decision making, the adaptation of machining conditions, or planned interventions. Fault detection and diagnosis models are used to monitor and analyze vibration signals in different configurations and conditions of use [21,22,23]. The variety of spindle types, the complexity of working conditions and testing conditions, and the diversification of failure modes lead to an abundance of mixed information, which is difficult to analyze, making diagnosis inaccurate or very complex [23].

The intensification of the field of mechanical processing in view of increasing productivity and precision, through the use of CNC machine tools, has necessitated equipment with spindles of high and very high speeds, up to 150,000 RPM or even greater in the case of grinding machines [24].

The purpose of this article is to propose an original approach to diagnosing a HSM spindle in the case of commissioning testing on grinding machines. The testing conditions are industrial—namely, monitoring in machine tool conditions—where the fulfillment of functions and functional limits are essential. In the machine tool, the conditions are different from those of a test rig; stiffness is distributed differently than on the bench, and lubrication and cooling are centralized, having specific local parameters. The frequency converter can also undergo certain modifications or optimizations from machine to machine. At the same time, the spindle can be new or repaired, but the functional and quality parameters must be respected under the same conditions. This work deals with commissioning monitoring for both, a new and a repaired spindle, determining the qualitative and quantitative parameters specific to vibrations, with great attention to the thermal state of the bearings. An important asset for spindle bearings’ condition is monitoring based on synchronous envelope analysis using the Hilbert transform [25,26].

One of the most widely used methods for bearing diagnosis is vibration enveloping. This technique is based on the structural characteristics of bearings and highlights the presence of shocks and friction from the early stages of defects. In the last decade, a series of signal analysis and processing methods based on the envelope method have been developed, but not all of them are used in industrial conditions, directly on machine tools, particularly in the case of spindles in high-speed grinding machines. Some research proposes new envelope techniques based on the instantaneous amplitude (envelope) and instantaneous frequency to obtain three novel envelope representations: instantaneous amplitude–frequency mapping, instantaneous amplitude–frequency correlation, and instantaneous energy–frequency distribution [21,22,23]. There are also studies that highlight the importance of methods in the dynamic evaluation of spindles, namely for evaluating the degradation over time of the vibration performance of bearings in machine tool spindles in a real-time monitoring framework, but the examination speeds are relatively low [22]. However, the use of evaluations based on advanced signal processing methods remains a certainty for the future [27]. Various sensors are used in spindle monitoring and diagnosis, in association with signal processing techniques such as the wavelet transform method, to decompose the vibration signal obtained using a vibrometer (laser Doppler vibrometer), but in conditions of up to 12,500 RPM [23].

In the case of grinding machines, spindle replacement is much more frequent due to processing at high speeds but also to the very high precision [28]. Thus, in-depth and precise knowledge of the spindle condition at commissioning is essential. High speeds generate intense friction, heat, and variations in electrical parameters, as well as high centrifugal forces [29,30]. In this context, geometric precision and the quality of dynamic balancing are essential. During the spindle run-in test, many catastrophic events can occur, such as the complete destruction of the bearings or compromise of the rotor or stator, which can lead to total failure or the need for subsequent repair [18]. Both of these options involve very high additional costs, as well as the loss of time.

This research proposes the development of a complex method for the monitoring of the spindle using a series of advanced signal processing techniques measured via high-speed acquisition modules. A key element in the proposed method is the determination at any time of the dynamic condition of the spindle. In the following section, a real-time monitoring and diagnostic method is proposed for industrial applications when commissioning high-speed spindles.

2. Monitoring and Diagnosis Method

The need for very high-speed machining (HSM) and high precision in grinding processes is well known, as is the need for high-speed spindles. The method used in this research consists of advanced monitoring and vibration analysis in the case of HSM spindle commissioning on grinding machines. During spindle commissioning, testing the running condition is necessary in order to evaluate the state of the spindle for the grinding process. In many situations, spindle replacement is a challenge, especially in the case of high-speed spindles, requiring a specific protocol. Many scholars have approached vibration and thermal monitoring from the perspective of determining certain types of defects in laboratory conditions [6,17,31], but industrial conditions are different, with a multitude of external factors interfering [32,33,34]. In this context, we propose a monitoring method using the monitoring capabilities offered by both the machine and additional sensors, as shown in Figure 1. The spindle can be either new or repaired, but the monitoring and evaluation parameters are the same.

The main challenge during the run-in is the real-time monitoring of the parameters so that the risk of failure can be eliminated, the most important being a bearing failure, which can occur both on the test rig and on the machine tool. To perform bearing run-in during commissioning, it is imperative that the temperature and electrical parameters, such as the power and electrical current, be monitored at the same time as vibration monitoring. As shown in Figure 2, the commissioning of the spindle requires a running test, with the speed increasing from the minimum speed to the maximum speed (tracking analysis).

During the increase in speed, the spindle temperature generally rises due to increased friction in the bearings and motor. This temperature increase can be significant, especially at higher speeds, and can affect the bearing performance [33,35]. Monitoring the temperature during the run-in test is crucial in maintaining optimal conditions. After reaching a stable temperature, the run-in test with vibration monitoring can be continued.

The proposed method considers the analysis of defects generated at the primary frequency domain and at the secondary frequency domain in order to identify the main categories of defects specific to spindle vibration diagnosis.

The originality of this research consists of the application of a real-time monitoring method based on the coupling between the frequencies and amplitudes of vibration obtained during testing on the machine. The innovative nature of the research is highlighted by the numerical development of an advanced signal processing algorithm based on the synchronous analysis of the envelope in both the linear and non-linear domains. This method presents the capacity to evaluate the fault state by monitoring the vibration level as a standard limit or a predefined threshold for envelope acceleration.

The method used in this research refers to advanced monitoring and vibration analysis. The synchronous envelope vibration analysis (SEVA) method is based on the Hilbert transform [25,26], seeking to identify mechanical defects and obtain a better response from the bearings. Envelope vibration is a technique used to extract the modulation signal to understand fault behaviors that occur at a high frequency. During a speed increase, envelope acceleration ensures the identification of bearing defects and thus any malfunction can be avoided. At high speeds, the centrifugal force is very high, generating excessive stress on the bearings, and the most affected element in these conditions is the cage. Due to the high acceleration because of the increase in speed, the cage can crack, potentially leading to imminent bearing damage.

A significant source of noise and vibration in high-speed machine tool spindles is the bearings, being subjected to high friction and temperatures, critical speeds, chatter, electrical variations, or high centrifugal forces.

The method consists of the real-time identification of the spindle condition through the monitored signals both in the low-frequency domain (imbalance, mechanical weaknesses, resonance, poor lubrication, electrical problems, etc.) and in the high-frequency domain (bearings, converter frequencies, etc.). The method presents the capacity to evaluate the fault state by monitoring the vibration level as a standard limit or a predefined threshold for gE acceleration.

Figure 3 shows the real-time monitoring system, which includes the first step concerning signal acquisition by sensors and transducers, followed by signal processing and analysis. The system is divided into two stages: the behavior at the main frequency stage, i.e., at low frequencies, and the dynamic condition at a high frequency. In the first stage, the dynamic behavior is analyzed to identify the high-energy vibration. In the second stage, the signals are processed via a synchronous fast Fourier transform (FFT), and resonance frequencies are filtered, followed by the Hilbert transform (HT) and then returning to time via the inverse fast Fourier transform (IFFT).

Then, the SEVA method in both the time and frequency domains is used to identify the critical impact frequencies of the bearing spindle. Moreover, the harmonics and interharmonics are highlighted to determine their distribution and repeatability. Finally, there is the identification of defects and the bearing condition [25,26]. The Hilbert transform $\widehat{x}\left(t\right)$ of a real-time $x\left(t\right)$ is defined as

$$\widehat{x}\left(t\right)=H\left(x\left(t\right)\right)={\displaystyle \frac{1}{\pi }}{\int }_{-\infty }^{\infty }{\displaystyle \frac{x\left(\tau \right)}{t-\tau }}d\tau$$

(1)

where the signal $x\left(t\right)$ is the impulse response function of the Hilbert transform [3]. The method to compute the Hilbert transform (HT) of a function, called the Hilbert transform function (HTF), applies to the frequency domain. In this case, the HT of $y(t)$ is $\widehat{y}\left(t\right)$ [25,26]:

$$\widehat{y}(t)={\int }_{-\infty }^{\infty }{\displaystyle \frac{y\left(\tau \right)}{t-\tau }}d\tau$$

(2)

The HT in the frequency domain is defined using $Y\left(f\right)$ and $\widehat{Y}\left(f\right)$, representing the Fourier transform (FT) of $y\left(t\right)$ and $\widehat{y}\left(t\right)$, i.e., $Y\left(f\right)$ and $\widehat{Y}\left(f\right)$, as follows:

$$Y\left(f\right)={\int }_{-\infty }^{\infty }y\left(t\right)exp-j2\pi ftdt$$

(3)

$$\widehat{Y}\left(f\right)=-jsgn\left(f\right)Y(f)$$

(4)

Applying the Fourier transform (FT) to the convolution defined in Equation (4), we can obtain [26]

$$\widehat{Y}\left(\overline{j\omega }\right)=FT\left[\widehat{y}\left(t\right)\right]=FT(1/\pi t) ·Y\left(\overline{j\omega }\right)=-jsgn\left(\overline{\omega }\right)Y\left(\overline{j\omega }\right)$$

(5)

The identification of the defect using the SEVA method is based on the HT to ensure good accuracy in the amplitude and on a complex algorithm for digital resampling to ensure high accuracy in the frequency [26]. For a better understanding of the method, a flowchart is presented in Figure 4. The envelope of the vibration signal is a low-frequency signal that follows the peaks of the rectified input signal.

The first signal processed is the acceleration, as shown in Figure 3 and Figure 4, where the frequency analysis is divided into two frequency domains: the low-frequency domain (LFD) and high-frequency domain (HFD). The first step in signal processing targets the vibration velocity parameter, after applying a low-pass filter (LPF), and corresponds to the identification and quantitative localization of the vibration energy generated by specific defects. However, frequency demodulation in the case of high-speed spindles takes place starting from a very high frequency range, applying a high-pass filter (HPF), in which band-pass filtering for resonant frequencies is applied (Figure 4). After obtaining the envelope spectrum, components with a frequency equal to the rate of occurrence of the impacts and with an amplitude proportional to their energy are identified.

The Hilbert transform in the time domain is subjected to the Fourier transform to obtain the demodulated frequencies after applying the band-pass filter. For the industrial application of the method in the case of spindle commissioning, an experimental protocol for high speeds is developed so that it is possible to track defect frequencies during speed increases (Figure 4).

3. Experimental Protocol and Signal Parameters

In order to implement the conditions to obtain specific monitoring signals according to Figure 3 and Figure 4, a test protocol is designed based on a general measurement procedure (Figure 5)—specifically, on a configuration to set the parameters necessary for acquisition and advanced signal processing.

3.1. Experimental Setup and Procedure

To understand the dynamic phenomena of the CNC grinding machine spindle, the present method consists of the correlation of vibration and mechanical actions in an experimental manner. To meet the acceptability conditions, it is necessary to choose the appropriate number of sensors. The choice of sensors is an important decision, considering the parameters that need to be monitored, the speed range, and the type of spindle. A complex experimental protocol is designed and implemented to highlight the three-dimensional vibration behavior of the spindle during running-in (Figure 5). The running-in of a spindle, both new and repaired, is essential for the stable dynamic behavior of the bearings and for the lifespan of the spindle. Taking into account the type of spindle, in most cases, the sensor mounting is applied only to the front bearing, and the minimum number of sensors required is two in the case of vertical spindles (one radial, X or Y, and another axial, Z) and three in the case of horizontal spindles (two sensors in the radial direction, X and Y, and the third in the axial direction, Z). When there is a structural stiffness problem between the spindle housing and the spindle motor, additional sensors can be used. In the case of very high-speed spindles, the spindle surface is small, and a three-dimensional accelerometer is used to optimize the number of sensors (Figure 5). For a deeper understanding of the structural behavior, an additional accelerometer is used and fixed in the radial direction, 1D accelerometer mounted in y direction (1D−y). The signals are synchronized with the speed, obtained using a laser speed sensor (Figure 5). The signals are acquired via a National Instruments USB4431 acquisition module, with 5 input channels (four vibration inputs and one speed input); signal processing and SEVA analysis are performed via the Fastview 17 software.

Figure 6 and Figure 7 show two examples of monitoring the running-in phase during commissioning, namely a horizontal spindle configuration and a vertical one, respectively.

The monitoring procedure is conducted during a speed increase by steps of 5000 RPM, starting from 10,000 RPM. In the case of high-speed spindles, lubrication is achieved by an oil/air mist, ensuring the specific lubrication conditions for high speeds. Additionally, stator cooling is another mandatory condition to be respected so that the temperature difference between the inlet and outlet of the cooling liquid does not exceed 4 °C. These conditions are tested before starting spindle rotation. During the running test, the casing temperature is also constantly monitored and must not exceed 50 °C. Vibrations are measured using a unidirectional accelerometer (1D accelerometer) of the Bruel & Kjaer type for high frequencies of up to 25 kHz, fixed in the Y direction (Figure 6 and Figure 7). For three-dimensional vibrations, in the current case, a three-dimensional accelerometer (3D accelerometer) is also used, so that the dynamics of the spindle in a three-dimensional configuration can be determined. The vibration signal directions according to the 3D accelerometer are as follows: the X direction represents the radial direction or the cutting speed direction, Y corresponds to the axial feed, and Z corresponds to the radial or depth of cut direction, as shown in Figure 6 and Figure 7.

In order to synchronize the vibrations with the speed, a laser speed sensor is used. Experimental monitoring is carried out under industrial conditions, directly on the machine. Because, in the industry, a specific optimization must be performed to reduce the necessary monitoring time, determine the minimum number of sensors, and identify the data extraction conditions, experimental tests are performed under reference conditions, when the spindle is in a new state.

3.2. Signal Parameters and Processing Data

Considering the SEVA method, as well as the experimental procedure under machine conditions, the signal processing and acquisition parameters are described so as to reflect the key points of the application.

Considering the case of a grinding machine with a high-speed spindle, and the requirement for high accuracy in the cutting process, the setting of the signal parameters is essential. Since the measurement protocol is focused on the commissioning of the main spindles, for the basic optimization of the monitoring parameters, tests were carried out directly on a HSM spindle. This experimental approach aimed to highlight the signal parameters so that the spindle could be characterized without the interference of the grinding parameters. The most important parameters used in the case of high frequencies are presented in

Table 1. Monitoring signal parameters.

RPM	Sample Rate	Buffer Size (Samples)	Block Size (Samples)
	(Samples/S)
Max. 10,000	25,000 (12 kHz)	32,768 (12,800 lines)	10,000
Max. 90,000	50,000 (25 kHz)	65,536 (25,600 lines)	25,000

.

To comply with these parameters, the signal acquisition and processing system has the following characteristics: a sampling rate of 100 kS/s/ch, a resolution of 24 bits, and an analysis buffer of 524,288 samples. This provides the characteristics necessary for a high-accuracy diagnosis. According to the speed range, being greater than 10,000 RPM, the datasets are extracted under the same operating conditions, with a sampling rate of 50,000 samples/s, a buffer size of 65,536 samples, and a block size of 25,000 samples. For high-speed grinding spindles (20,000–100,000 RPM), the suggested cutoff frequencies for envelope demodulation are as follows: low cutoff—1000–2000 Hz; high cutoff—25,000 Hz. These bands target the resonant amplification of small impacts caused by bearing defects, which are often masked in raw vibrations but become visible in the envelope. The fault frequencies caused by the occurrence of shocks due to the bearing defect depend on the speed and the geometry of the bearing. The frequency range for bearing condition monitoring (BCM) is determined by the defect coefficients of bearing elements: the ball pass frequency inner-ring (BPFI), ball pass frequency outer-ring (BPFO), ball spin frequency (BSF), and fundamental train frequency (FTF). For a better understanding, an example of signal processing according to the SEVA method is presented, highlighting the main analysis signals, in Figure 8, Figure 9 and Figure 10.

Acceleration remains one of the benchmark parameters in the field of vibration monitoring, providing the primary analysis characteristics (Figure 8). As described (Figure 3 and Figure 4), the analysis is divided into two configurations: one specific to the LFD, where imbalance, misalignment, lubrication, or electrical defects are highlighted (Figure 9), and one for the HFD to monitor the condition of the bearings (Figure 10). By applying the Hilbert transform (HT) to the filtered signal, the signal envelope (i.e., the magnitude of the analytic signal) can be obtained. This time-domain envelope reveals the periodic impacts caused by faults, and its frequency spectrum can then be analyzed to identify bearing defects (Figure 10).

In the case of motor spindles, the analysis can be simplified due to the compact main shaft/rotor/stator assembly. However, the appearance of difficulties is replaced by dynamic and thermal phenomena that occur at high speeds. The quality of the acquired signals is influenced by the quality of the sensors and the choice of monitoring parameters. As shown in Figure 9, at 50,000 RPM, the fundamental frequency 1X is present and monitored, providing in real time the level of imbalance. At the same time, at very low frequencies, the air pressure generated by the drainage of the air/oil lubrication system is present. This information is also taken into account, but it is filtered out when the specific identification of the faults takes place, as shown in Figure 8.

Knowing the resonant frequencies, a band-pass filter with multiple filtering ranges can be used to determine the specific frequency range for bearing condition monitoring (BCM). Figure 10 shows the enveloped acceleration spectrum for a speed of 50,000 RPM, after filtering the frequencies of interest in the range of 2000–25,000 Hz. The existence of harmonics from 1X to 6X can be observed, but without an impact on the condition of the bearings. Non-harmonics are also present at high frequencies, which are specific to monitoring the defects of the bearings.

4. Results and Discussion

During commissioning, the monitoring of the vibration parameters and also the temperature is necessary. A bearing failure may be imminent, but, at very high speeds, the occurrence of other damage is also likely. The signals from in situ conditions contain more noise and vibration interferences than in test rig conditions, the measurement environment is complex, and it is necessary to take into account local factors. In the following, some specific cases of commissioning spindles are presented.

The proposed method was used in various cases of the commissioning of grinding machines, where the spindle was repaired and subjected to running-in testing. Spindle repair can be standard, such as replacing bearings and seals and other normal repair operations, or non-standard, which involves the machining of important components, the metallization of surfaces, stator rewinding, or rotor repair, etc. A lack of monitoring in the case of grinding machines can lead to the failure of some spindles, as shown in Figure 11. However, repair can also be carried out at this level, but specific monitoring is crucial during commissioning. Under certain conditions, real-time evaluation can also enable the identification of criteria that impose speed range limitations.

After repairing the rotor and rewinding the stator, the spindle was balanced according to ISO21940 G0.4 [36] and mounted with specific bearings, i.e., GMN HY SM6001 (front bearing) and GMN HY SM6000 (rear bearing), finally respecting the imposed geometric conditions, with a maximum radial run out of 1 μm and maximum axial run out of 1 μm.

In industry, we can encounter different cases in which the vibration velocity is high and the acceleration level is low (the existence of a defect at low frequencies with high energy, independently of a bearing defect) and vice versa (the existence of a bearing defect in an incipient stage) or in which both a high vibration velocity and high vibration acceleration are presented—an interdependence between acceleration and velocity, indicating that the bearing defect is in an advanced stage of wear. An advanced bearing failure will generate high energy vibrations sensitive to the vibration velocity. In this paper, the focus is on the real-time monitoring and evaluation of the spindle during commissioning, and the clear identification of each component is essential. During commissioning, low-frequency faults must be known so that they can be differentiated from those occurring at high frequencies, which are specific to bearing failures.

The use of real-time monitoring and analysis for operating spindles is essential to determine the operational status and bearing condition and prevent premature wear; in this context, the vibration velocity and vibration acceleration parameters are necessary (

Table 2. Practical frequency thresholds for high speed grinding spindles.

Domain	Frequency Range	Common Faults Detected	Parameter
Low Frequency	1–2000 Hz	Imbalance Misalignment Lubrication Looseness issues Resonance Electrical defect Cooling issues	Velocity vibration (mm/s rms)
Mid Frequency	1–5 kHz	Gear mesh Bearing faults Impacts Resonance	Acceleration (g rms) Envelope acceleration (gE rms)
High Frequency	5–30+ kHz	Bearing faults Impacts Converter faults

). In the demodulated frequency domain, the standards ISO 20816 [37], ISO 13373-1 [38], and ISO 15243:2017 [39] do not specify a frequency range, but, in practice, these ranges are typically 1–30+ kHz for high-speed spindles [40,41,42,43]. Practical frequency thresholds for grinding spindles’ vibration monitoring are presented in Table 2.

4.1. Vibration Velocity Monitoring

The vibration velocity parameter is used for the overall evaluation of the spindle, being the most frequently used criterion during commissioning, and is also included in the ISO standards.

The parameter of the vibration velocity is very important and decisive, but it cannot cover the whole range of defects. In the case of a bearing defect, the vibration velocity cannot be used to detect the defect in its early stages, as it is only determined in the low-frequency range, at the advanced stage of a defect.

Harmonic frequencies are investigated to evaluate the spindle behavior from the point of view of low-frequency phenomena (see Figure 12). The 1X harmonic is the main factor that can indicate the uniform movement of the spindle and thus the quality of the balancing, where Vrms represents the overall velocity vibration. The first case studied is the GMN TSSV100S spindle with a maximum speed of 90,000 RPM, with 3 kW, 350 V, 7.7 A, and 1500 Hz. The spindle was subjected to run-in after a non-standard repair, as shown in Figure 12.

In Figure 12, the evolution of the vibration velocity can be observed as a function of the speed, with a significant increase in amplitude at a speed of 70,000 RPM, which also represents a speed limitation in this case. The amplitude of the vibration velocity increases with increasing speed, reaching the limit of 1.8 mm/s rms at a speed of 70,000 RPM, marking a critical operating point for this spindle. The frequency distribution at 70,000 RPM is shown in Figure 13, where the 1X amplitude is high, reaching 1.43 mm/s rms and an overall level of 1.9 mm/s rms. According to ISO 21940-1 [36] and ISO20816-1 [37], the amplitude of the fundamental frequency increase exceeds the limit, and, to avoid premature wear, the speed limit is set to 70,000 RPM, resulting in the optimal cutting speed being achieved by increasing the diameter of the grinding tool.

The use of repaired spindles is common across all CNC machines, especially in the case of spindle motors, where the compatibility of these components allows for quick replacement. Many machine tools do not have a real-time monitoring system, and the use of the vibration velocity is crucial for high-speed spindles to ensure optimal performance and prevent potential damage. Monitoring the vibration velocity provides valuable insights into the overall health of the spindle system, allowing for proactive maintenance and preventing unexpected breakdowns, as well as ensuring the optimal conditions for the correct prediction of the machining process [44].

The velocity monitoring parameter is also applied to another spindle, a Fischer type with 90,000 RPM, 2.5 kW, 350 V, 6 A, and 1500 Hz (see Figure 14). In this case, the spindle was subjected to standard repair, with the evaluation method being the same. We observe the evolution of the overall vibration velocity and the first-order harmonic, respecting the limit of 1.8 mm/s rms up to the maximum speed of 90,000 RPM.

In the case of this spindle, the commissioning took place under optimal conditions across the entire speed range, meeting both the vibration and temperature requirements. The frequency spectrum of the vibration velocity at 90,000 RPM shows stable behavior, where the amplitude of the 1X harmonic is 0.44 mm/s, in accordance with the imposed limits (see Figure 15).

As stated above and in the literature [40,41], the vibration velocity cannot ensure safe and accurate monitoring at high frequencies. For a correct assessment of the bearing fault’s condition, acceleration enveloping is needed.

4.2. Acceleration Envelope Monitoring

According to the online monitoring system, the second stage is to monitor the condition of the bearings during the running-in period. As stated earlier, the risk of bearing deterioration during a speed increase is high. At high speeds, due to very high accelerations, dynamic and thermal phenomena can generate premature wear in the bearings, and the most common case is cage failure.

Enveloped acceleration is a signal processing technique used in vibration analysis, particularly for detecting early signs of damage in bearings and gearboxes. It works by filtering out the overall vibration signal and highlighting the repetitive impacts caused by defects. Figure 16 shows the evolution of the acceleration—specifically the enveloped acceleration—in the case of the GMN spindle with non-standard repair. It is necessary that bearing condition monitoring be performed at a high level to enable the real-time detection of defect occurrence. The evolution of bearing failures is exponential, and knowledge of the specific defect frequencies is essential. In comparison to the vibration velocity, the acceleration envelope is stable and within the acceptable range, even at high speeds, but not beyond 72,000 RPM. The frequency spectrum shows the presence of the first three harmonics, but, at the speed of 72,600 RPM, the frequencies responsible for the bearing condition are present. The excitation of bearing frequencies is caused by faults generated by the first-order harmonic (1X). The spindle in question underwent a complex repair, and monitoring during the spin test was crucial. In the case of commissioning, it is essential to identify the critical points that provide sufficient data for decision making. As shown in Figure 16, it is important to follow the evolution of vibration acceleration during spindle commissioning. The trend of acceleration is exponential, and, based on the SEVA method; it was possible to determine the critical operating point before the appearance of an accidental failure. Thus, the increase in vibration velocity (Figure 12) is generated by a rise in the bearing load under the conditions of a high speed and elevated centrifugal force. The speed limit range of the spindle can be observed. Acceleration data (Figure 16) complement velocity monitoring (Figure 12) very well, as the spindle speed is continuously varied. In the envelope acceleration spectrum (Figure 17), the emergence of defects is highlighted by both synchronous frequencies of orders 1, 2, and 3 and especially by non-synchronous frequencies related to the bearing condition, which excite the natural frequencies of the bearing elements.

The importance of the SEVA method is highlighted by its ability to provide the high-frequency data required for root cause diagnosis. Thus, the location of the critical speed point at 72,600 RPM is due to excitation in the bearings and not to any specific velocity vibration defect.

The same procedure regarding envelope acceleration is performed for the Fischer spindle at start-up. Up to 90,000 RPM, the maximum vibration level was 2.7 gE at 70,000 RPM, and, at 90,000 RPM, the envelope acceleration level reached 2 gE; see Figure 18. The envelope acceleration spectrum (see Figure 19) is centered on harmonic frequencies synchronous with the rotational speed. In contrast, the specific BCM-type frequencies show reduced amplitudes, indicating the good operating condition of the bearings.

During commissioning, a series of factors must be considered to ensure that monitoring provides specific parameters for the evaluation of the spindle condition in real time. This research aims to highlight the importance of real-time monitoring using innovative and direct diagnostic methods, seeking to avoid breakdowns and accurately evaluate the operational state of the spindle.

5. Conclusions

This research has introduced and validated an advanced method for the real-time monitoring and vibration diagnosis of high-speed spindles during commissioning on grinding machines, aiming to ensure optimal operational performance and prevent defects. The proposed method is centered on synchronous envelope vibration analysis (SEVA), integrating both low- and high-frequency domain data, and leverages Hilbert transform-based signal demodulation to identify critical bearing and mechanical faults. Synchronous envelope vibration analysis (SEVA) combines two well-known methods: envelope analysis, which is a method that demodulates high-frequency signals to detect impacts from defects like cracks or bearing defects, and synchronous signal processing, which aligns the vibration signal with a reference (usually a shaft rotation or gear mesh period) to better isolate periodic components. The novelty of this paper lies in the integration of the SEVA method with a signal filtering technique to enable bearing condition monitoring (BCM) during the commissioning of grinding machines. Additionally, the approach can also detect other defects, including those associated with the frequency converter.

This study highlights that the commissioning stage is a critical operational phase, particularly for high-speed spindles, where mechanical and thermal stress can generate significant failure risks. Monitoring during this phase allows the early detection of anomalies such as imbalance, poor lubrication, resonance, or bearing degradation.

The dynamic characteristics determined in the case of the two spindles also represent a reference for the statistical data necessary to evaluate the behavior of the spindle in the context of predictive maintenance. In this work, two types of very high-speed spindles (90,000 RPM) were analyzed under critical conditions during commissioning. Both spindles performed optimally during the run-in phase with a speed increase, even though one of them underwent a non-standard repair, which involved not only the replacement of bearings, seals, preloading, balancing, etc., but also the repair of the rotor by metallization and stator rewinding. Considering these challenges, the SEVA method proved effective in ensuring qualitative and quantitative control over the vibration parameters, and the run-in took place under optimal conditions.

The real-time acquisition and processing of vibration and rotational speed signals, combined with the dual-domain analysis framework, provides a comprehensive diagnostic approach.

Key findings and contributions of this research include the following:

• The development of a dual-stage monitoring protocol, combining low-frequency domain analysis (LFD) to detect structural and mechanical faults with high-frequency domain analysis (HFD) for the precise evaluation of bearing conditions.
• The implementation of the SEVA method, a vibration signal processing algorithm that uses a synchronous FFT, the Hilbert transform, and envelope analysis. This technique allows for high sensitivity in detecting cage and rolling element defects in bearings at early stages.
• The establishment of experimental parameters suitable for industrial application, including sampling rates, filtering ranges, and sensor configurations (1D and 3D accelerometers), all optimized for the dynamic behavior of high-speed spindles.
• The identification of critical vibration patterns, including both harmonic and non-harmonic components, with the capability to differentiate between normal and defective conditions, even under high-speed operation.
• Practical validation under industrial conditions, which proves the robustness and adaptability of the proposed method in real-world manufacturing environments, beyond laboratory simulations.

The originality of this research lies in the integration of synchronous envelope vibration analysis with real-time signal acquisition and processing, forming a complete methodology applicable to both new and repaired spindles. Unlike conventional offline or laboratory-based methods, the proposed approach ensures in situ diagnosis with reduced downtime, contributing significantly to predictive maintenance strategies in high-speed machining contexts. Spindle measurements and analysis are carried out under two scenarios: during normal operation and during commissioning. For an operational spindle, the process typically takes around 1 h for a standard machine, assuming that all necessary tools are available and properly configured. The procedure involves several steps: sensor installation and cabling, tachometer/encoder setup, signal acquisition, data pre-processing (including filtering, the Hilbert transform, and synchronization), SEVA analysis (envelope FFT and interpretation), and final report generation. In the case of spindle commissioning, the required time depends on multiple factors, such as the spindle type, lubrication system, bearing type, cooling system, tool clamping mechanism, and operating speed. On average, the monitoring and diagnostic process for a high-speed grinding spindle takes approximately 6 to 7 h.

Subsequent development will involve integrating a machine learning model to establish a numerical monitoring framework with the ability to autonomously identify and characterize defects.