**1. Introduction**

Plasma electrolytic oxidation processing allows the creation of durable, thick, uniform and strongly adherent coatings on valve metals [1,2]. Extensive investigations on plasma electrolytic oxidation have been carried out in past years, due to its increasing use in industrial applications. Most of the research focuses on processing conditions [3–8], discharge characteristics [1,9–13], coating microstructure [14–16], mechanical properties [17], environmental performance [16,18], and functional characteristics [2,18]. As a result, the correlations [3–5,13–15,19] between the electrical conditions, the electrolyte compositions, the coating microstructure, and the growth rate of coating, which are linked via the characteristics of the discharges, have become clearer over recent years. Moreover, progress has been made [20–25] in unraveling the inter-relation between the energetics of individual discharges, the pulse energy, the soft-sparking regime, and the energy consumption.

Recently, the main research focus has been on the dielectric breakdown of oxide films and on the associated discharges that repeatedly occur over the whole surface of the metal substrate. It is clear that most of the new oxide of the coating created during each discharge is formed within the plasma as it cools and collapses [1,5,26]. Although the nature of the plasma created via the discharges, which occur during the plasma electrolytic oxidation process, is still uncertain, it is clear that the discharge characteristics are affected by a series of electrical conditions. Hence, there is considerable scope for more effective process electrical control, with specific objectives in terms of coating performance and energy efficiency, and an attempt is made to identify key points that are likely to assist this.

Several common electrical conditions, such as the use of direct current (DC), alternating current (AC), unipolar, and bipolar pulses, have been employed over recent years. It has been repeatedly found that the AC mode can provide more e ffective processing and higher quality coatings [1,27]. Furthermore, square waveforms and higher frequencies are being increasingly exploited in research and in commercial use, due to their more sophisticated impulse energy control capabilities [28–30]. However, the most common control modes at the basis of the plasma electrolytic oxidation processing can be classified as current-control and voltage-control. Meanwhile, most of the electrical parameters, which are applied in practice, are fixed or are chosen within a range of preset values on a largely empirical basis [31]. The major disadvantage of such control modes is that the applied electrical parameters cannot be adjusted automatically to match the dynamic requirements of the plasma electrolytic oxidation coating during its growth process.

Furthermore, high energy consumption restricts the performance of traditional plasma electrolytic oxidation processing; by exposing a metal substrate to oxidizing agents and highly energetic discharges, in fact, its discharge mechanism is extremely ine fficient. Troughton et al. explored the energies absorbed by various phenomena taking place during a plasma electrolytic oxidation process (melting and vaporization of substrate, melting of existing oxide coating, initiation and sustaining of the plasma, vaporization of water, and electrical heating of the electrolyte) and inferred that most of the injected energy is absorbed in the form of the vaporization of water [32]. What is more, there are several studies [1,24,25] aimed at high energy e fficiency and helpful measures. The most promising approach is probably to somehow reduce the energy associated with each discharge, possibly by reducing the voltage needed for it to occur, or to promote more transformation per discharge. Therefore, it is urgen<sup>t</sup> to develop an energy-e fficient intelligent process control method.

In this study, a novel self-adaptive control method based on a real-time diagnostic and a precision pulse energy regulation technique was employed for the plasma electrolytic oxidation processing of 6061 aluminum alloys. Moreover, the electrical characteristics, coating microstructure, and energy consumption of such processes were investigated.

#### **2. Materials and Methods**

## *2.1. Materials*

A quantity of 6061 aluminum alloy samples with dimensions of 60 mm (L) × 60 mm (W) × 2 mm (H) and a 2 L water-cooled stainless steel tank, which served as a counter-electrode, were used. KOH (5 g <sup>L</sup>−1) and Na2SiO3 (10 g <sup>L</sup>−1) were dissolved in distilled water to form the electrolyte. The plasma electrolytic oxidation process was performed by using a home-built 20 kW unipolar pulsed power supply, working in voltage-control mode. The supply could provide a maximum voltage of 600 V and a maximum frequency of 10 kHz. The temperature of the setup was maintained at 25 ◦C.

#### *2.2. Self-Adaptive Control Method*

The applied self-adaptive control model was operated in a double-closed loop (Figure 1a). The voltage loop ensured a constant-voltage control, whereas the voltage–current loop ensured the online adjustment of the process parameters via an active diagnostic method. The complete workflow of the self-adaptive plasma electrolytic oxidation process investigated in this paper is presented in Figure 1b. Initially, the samples were anodized up to 300 V at a rate of 20 V/min. Successively, initial test pulses were applied to determine the initial breakdown voltage ( *Vi*) and the termination voltage (*Vu*) of the sample. These parameters are listed in Table 1 (type 1). Finally, the plasma electrolytic oxidation processing was implemented by using the self-adaptive unipolar voltage pulsed mode (*f* = 100 Hz, *d* = 50%). The applied voltage was dynamically adjusted by using Equation (1):

$$
\mathcal{U} = V\_b + \Delta \mathcal{U}.\tag{1}
$$

Here, *U* is the voltage applied to the sample, *Vb* corresponds to the breakdown voltage, and Δ*U* represents the voltage deviator (0–5 V).

**Figure 1.** Process control method: (**a**) Self-adaptive control model, (**b**) self-adaptive control process, and (**c**) feature information identification algorithm.


**Table 1.** Parameters obtained during the test pulses experiment.

\* In type 2, the step number of test pulses is equal to 5, the pulse number for the same step is equal to 5, and *Vb*<sup>n</sup>−<sup>1</sup> corresponds to the last breakdown voltage.

The processing times to maintain the same applied voltage magnitude and for completing the total process were determined by using the critical condition 1 (Equation (2)) and 2 (Equation (3)), respectively.

$$I \le \frac{1}{2} I\_b. \tag{2}$$

$$
\mathcal{U} \not\supseteq V\_{\mu}.\tag{3}
$$

In Equations (2) and (3), *I* and *U* represent the real-time feedback current and the voltage, whereas *Ib* corresponds to the breakdown current.

The key identification algorithm to obtain the coating feature information is shown in Figure 1c. The parameters of the test pulses acquired during the plasma electrolytic oxidation process are listed

in Table 1 (type 2). The identification criterion 1 was used to determine the dynamic values of *Vb* and *Ib* as follows:

$$\text{If } \left(\frac{di}{dt}\right)\_n \approx 0 \text{ and } \left(\frac{di}{dt}\right)\_{n+1} \ge 1, \text{ then } V\_b = V\_{n\prime} \ I\_b = I\_n.$$

The identification criterion 2 was used to determine the dynamic values of *Vu* as follows:

$$\text{If} \left(\frac{di}{dt}\right)\_n \ge 0, \left(\frac{di}{dt}\right)\_{n+1} \le 0, \text{ and } (I\_n > I\_b), \text{ then } V\_n = V\_{n+1}$$

where, *di dtn* is the rate of current change of pulse n, *di dtn*+<sup>1</sup> is the rate of current change of pulse *n* + 1, *Vn* represents the voltage magnitude of pulse n, and *I*n represents the current magnitude of pulse n.

A detailed description of the pulse test method and of the feature information extraction can be found in a previous work [6].

#### *2.3. Data Monitoring and Pre-Processing*

The voltage and the current data were detected by using a hall voltage sensor (CHV-25P) and a hall current sensor (CSM010B), and recorded with a sampling frequency of 1 MHz via a data acquisition card NI PCI-6133 controlled by LabVIEW software.

Considering the signal interference factors that exist in the plasma electrolytic oxidation process, such as mechanical vibration, electron avalanche, dielectric breakdown, and high-frequency electronic switching, a pre-processing of raw voltage and current data was carried out. The data were first filtered with a low-pass filter with a cut-off frequency of 50 kHz. This allowed high-frequency noise due to electronic switching to be eliminated. Next, these data were smoothed using a moving average filter. Finally, an appropriate scale function and wavelet basis function were adopted to extract a random noise signal.

#### *2.4. Post-Processing of Samples*

The surface morphology, the cross-section, and the chemical composition of the coatings prepared on the samples were observed by employing a LEO1530 VP scanning electron microscope (SEM) equipped with an X-ray energy dispersive spectroscopy (EDS) setup. The phase of the coating was analyzed via X-ray diffraction (XRD, Model D8 Advance) by using a Cu kα radiation source.
