Au-Pd Co-Modified LaFeO₃ for Low-Concentration n-Butanol Sensing and Composition Analysis of n-Butanol/HCHO Mixtures

Abstract

To improve the response of LaFeO₃ to low concentrations of n-butanol, LaFeO₃ co-modified with 3 wt% Au and 5 wt% Pd was prepared by a citrate-assisted sol–gel method. At 180 °C, the resulting sensor exhibited a response of 16.92 to 1 ppm n-butanol, corresponding to 5.90 times the maximum response of pristine LaFeO₃. The optimal operating temperature decreased from 200 to 180 °C, and a distinguishable response was observed at the lowest tested concentration of 0.1 ppm. However, selectivity tests revealed a substantial cross-response to formaldehyde (HCHO), motivating further analysis of binary n-butanol/HCHO mixtures. Responses to these mixtures were measured at 160, 180, and 200 °C. The ratio of the response increments at 160 and 180 °C reduced the influence of total concentration while retaining composition-related information. A support vector regression model with a radial basis function kernel was trained using the dataset obtained at a total concentration of 1.0 ppm. Without refitting, the model predicted n-butanol fractions for five predefined mixture compositions at total concentrations of 0.5 and 2.0 ppm. The mean absolute errors were 0.075 and 0.033, with R² values of 0.921 and 0.981, respectively. These results demonstrate the potential of combining Au-Pd modification with temperature-response normalization for low-concentration n-butanol sensing and binary-mixture composition analysis using a single LaFeO₃ sensor.

1. Introduction

N-butanol is widely used as an organic solvent in coatings, resins, and acid-cured paints and is also an important feedstock for the production of plasticizers and various fine chemicals. Owing to its high volatility, n-butanol vapor is readily released into occupational environments during production, coating, and film-curing processes. Exposure can irritate the eyes and respiratory tract and may even cause headache, dizziness, and central nervous system depression [1]. A field investigation involving 557 workplace air samples collected from 27 wood-product and furniture enterprises detected n-butanol in nearly 90% of the samples, with concentrations ranging from 0.02 to 32.6 ppm [2]. Recent studies on coating emissions have further confirmed that n-butanol can be released from waterborne coatings and is widely present in automobile- and ship-coating environments at concentrations ranging from sub-ppm to tens-of-ppm levels [3,4,5]. These findings indicate that n-butanol exposure arises not only from accidental high-concentration releases but also from its continuous low-level emission during routine production. Therefore, developing sensing materials capable of responding effectively to n-butanol in the sub-ppm to low-ppm range is of substantial practical importance for real-time tracking of concentration fluctuations, early warning, and exposure-risk control.

At present, n-butanol is mainly analyzed using gas chromatography, gas chromatography–mass spectrometry, and photoionization detection [6,7,8]. Although these instrumental methods offer high analytical accuracy, their equipment costs and operational requirements limit their implementation for low-cost, distributed, and continuous on-site monitoring. In contrast, chemiresistive metal oxide semiconductor gas sensors offer several advantages, including simple structures, low cost, ease of integration, and suitability for continuous monitoring [9,10,11]. Among metal oxides, ABO₃-type perovskites possess stable crystal frameworks, flexible A/B-site compositions, and tunable oxygen nonstoichiometry. These characteristics allow carrier transport, surface defects, and redox activity to be regulated while retaining the structural stability of the host lattice, providing considerable flexibility for optimizing gas-sensing performance [12,13,14]. LaFeO₃ is a typical p-type perovskite semiconductor with a stable FeO₆ octahedral framework, favorable thermochemical stability, and active surface oxygen exchange. Changes in its hole concentration induced by surface reactions can be directly converted into measurable resistance signals [15]. However, pristine LaFeO₃ provides a relatively limited accessible surface area and number of surface active sites, and its response toward low concentrations of n-butanol requires further improvement.

Noble-metal surface modification can effectively enhance the gas-sensing performance of ABO₃-type perovskites [16,17,18,19,20]. Noble-metal nanostructures can promote gas adsorption and surface reactions through catalytic activation, surface spillover, and interfacial charge regulation while amplifying the resistance modulation induced by surface reactions. Au nanostructures can enhance the surface adsorption of alcohol molecules and regulate interfacial charge exchange [17,20,21,22], whereas Pd/PdO_X species can promote the adsorption and activation of oxygen and alcohol molecules and improve reaction efficiency through surface spillover [18,19,23,24]. Single-noble-metal modification generally emphasizes either catalytic activation or interfacial electrical modulation, whereas Au–Pd co-modification can combine the interfacial regulation of Au with the catalytic activation of Pd, thereby strengthening both the surface reaction and electrical transduction processes [25,26].

N-butanol often coexists with formaldehyde (HCHO) and other volatile organic compounds (VOCs) in coating and furniture-manufacturing environments [2,27,28]. Because both n-butanol and HCHO can change the resistance of LaFeO₃ through reactions with surface-adsorbed oxygen, the response magnitude obtained at a single operating temperature cannot distinguish the respective contributions of the two components. Different gases exhibit distinct temperature dependences in their adsorption, activation, and surface-reaction processes. Therefore, combining multi-temperature responses with machine learning provides a feasible route for analyzing mixed-gas compositions using a single sensing material [29,30,31,32,33,34,35].

Based on this rationale, 3 wt% Au and 5 wt% Pd were simultaneously introduced into LaFeO₃ following single-metal loading optimization (Figure S1) to prepare Au-Pd/LaFeO₃. The material exhibited a response of 16.92 toward 1 ppm n-butanol at 180 °C, approximately 5.90 times the maximum response of LaFeO₃, while reducing the optimal operating temperature from 200 °C to 180 °C. It also produced a distinguishable response at the lowest tested concentration of 0.1 ppm. BET and XPS results indicate that Au-Pd/LaFeO₃ possesses a larger accessible gas–solid interfacial area and higher proportions of defect-related and adsorbed-oxygen-related surface components, which may provide favorable conditions for the adsorption and surface reactions of oxygen and n-butanol. To address the cross-response induced by HCHO, the responses of Au-Pd/LaFeO₃ to binary n-butanol/HCHO mixtures were further recorded at 160 °C, 180 °C, and 200 °C. A temperature-normalized feature F_160/180 was constructed using the excess response at 180 °C as the reference. The RBF-SVR model established using the 1.0 ppm dataset quantitatively resolved the n-butanol component at total concentrations of 0.5 and 2.0 ppm, yielding MAE values of 0.075 and 0.033, respectively. By enhancing the gas response through Au-Pd surface regulation and resolving cross-response signals through the temperature dimension, this integrated strategy enables a single LaFeO₃ material to perform both low-concentration n-butanol detection and binary-mixture composition analysis, providing a new route to reduce the dependence of VOC sensing systems on multi-material arrays.

2. Experimental

2.1. Material Synthesis

Au- and Pd-modified LaFeO₃ powders were synthesized using a citrate-assisted sol–gel method. Stoichiometric amounts of La(NO₃)₃·6H₂O and Fe(NO₃)₃·9H₂O were dissolved in 150 mL of deionized water at a La:Fe molar ratio of 1:1. Citric acid was subsequently added with a citric acid-to-metal-cations molar ratio of 1.5:1.

For samples modified with noble metals, calculated amounts of HAuCl₄·4H₂O and/or PdCl₂ were introduced into the precursor solution to obtain the required nominal Au and Pd loadings. The nominal noble-metal loading was calculated on an elemental-metal basis relative to the mass of LaFeO₃ host powder. The resulting solution was stirred for 6 h at 60 °C and then heated at 80 °C under continuous stirring until a homogeneous gel was formed. The gel was dried at 120 °C for 4 h and subsequently calcined in air at 700 °C for 3 h.

For preliminary optimization of the single-metal loadings, Au/LaFeO₃ samples containing 2, 3, and 4 wt% Au and Pd/LaFeO₃ samples containing 4, 5, and 6 wt% Pd were prepared under identical conditions. Based on the loading-screening results presented in Figure S1, 3 wt% Au and 5 wt% Pd were selected as the optimal single-metal loadings within the tested ranges. These two selected loadings were subsequently combined to prepare LaFeO₃ co-modified with 3 wt% Au and 5 wt% Pd.

Four representative materials were selected for systematic characterization and gas-sensing comparison: pristine LaFeO₃, LaFeO₃ modified with 3 wt% Au, LaFeO₃ modified with 5 wt% Pd, and LaFeO₃ co-modified with 3 wt% Au and 5 wt% Pd. These samples are hereafter denoted as LFO, Au/LFO, Pd/LFO, and Au-Pd/LFO, respectively.

2.2. Sensor Fabrication

The synthesized powders were mixed with deionized water at a powder-to-water mass ratio of 1:2 to form homogeneous slurries. A fixed amount of each slurry was deposited onto the outer surface of an alumina ceramic tube using a rotational coating process at 800 rpm for 0.5 h. The fabricated sensors were thermally aged in air at 200 °C for 18 h before gas-sensing measurements. All sensors were fabricated under identical conditions. The overall preparation process is illustrated in Figure 1.

Figure 1. Schematic illustration of the citrate-assisted sol–gel synthesis of LFO, Au/LFO, Pd/LFO, and Au-Pd/LFO and the fabrication of the corresponding gas sensors. The four materials were prepared without noble-metal precursors, with a Au precursor, with a Pd precursor, and with both precursors, respectively.

2.3. Material Characterization Methods

The phase compositions and crystal structures of the prepared materials were characterized by X-ray diffraction (XRD). Surface morphology and elemental distribution were examined using field-emission scanning electron microscopy (FESEM) coupled with energy-dispersive X-ray spectroscopy (EDS). The surface elemental composition and chemical states of Au, Pd, and O were analyzed by X-ray photoelectron spectroscopy (XPS). Nitrogen adsorption–desorption measurements were performed to determine the specific surface areas and pore-size distributions of the materials.

2.4. Gas-Sensing Measurements and Parameter Definitions

Gas-sensing measurements were performed using a WS-30A (Zhengzhou Winsen Electronics Technology Co., Ltd., Zhengzhou, China) gas-sensing test system equipped with an 18 L chamber. The target gases were supplied from standard gas cylinders, with air used as the dilution gas. Gas dilutions were prepared by mixing the standard gases with air at specified volume ratios using a syringe.

The relative humidity in the chamber was adjusted using a dynamic humidity controller and monitored in real time using an associated humidity sensor. After the chamber reached the specified relative humidity, the gas flow was stopped, and the target gas was injected for measurements in the closed chamber. Measurements were conducted at an ambient temperature of 20 °C. The relative humidity was maintained at 20% for the routine measurements and varied from 20% to 90% for the humidity-dependent tests.

Response time and recovery time are used to evaluate the dynamic characteristics of sensors. The response time refers to the time taken for the sensor resistance to reach 90% of its total variation upon introducing the target gas, while the recovery time represents the time required for resistance to return to 90% of the total change after switching back to air atmosphere.

The resistance of sensing layers is derived from the circuit relationship as:

$$R_{\mathit{sensor}}={\displaystyle \frac{V_c-V_{\mathit{out}}}{V_{\mathit{out}}/R_1}}$$

For the p-type sensing materials investigated herein, the sensor response (S) is defined as the resistance ratio under target gas and air atmosphere, as:

$$S={\displaystyle \raisebox{1ex}{$R_g$}\!\left/ \!\raisebox{-1ex}{$R_a$}\right.}$$

In the above expressions, R_g corresponds to the resistance recorded under target-gas exposure, and R_a represents the baseline resistance measured in ambient air.

2.5. Machine-Learning Analysis

For the analysis described in Section 3.3, the 1.0 ppm dataset contained 40 measurement records, with eight repeated measurements for each of the five mixture ratios. The 0.5 and 2.0 ppm datasets each contained 15 records, with three repeated measurements per ratio. The former was used for model development, whereas the latter two were used for concentration-transfer evaluation.

The F_160/180 values were standardized by z-score transformation within the model-development procedure. During cross-validation, the scaling parameters were obtained only from the corresponding training fold. The final scaling parameters were then determined from the complete 1.0 ppm dataset and applied unchanged to the 0.5 and 2.0 ppm datasets. Model parameters were selected using repeated four-fold cross-validation stratified by mixture ratio, with 25 repetitions. The final model used C = 30, ε = 0.03 and γ = scale, corresponding to γ = 1.0 for the standardized single-feature input. Predicted fractions were constrained to the range of 0–1 before error calculation.

3. Results and Discussion

3.1. Material Characterization

The XRD pattern of Au-Pd/LFO agrees well with the standard pattern of orthorhombic LFO (JCPDS No. 37-1493). All major diffraction peaks can be indexed to LFO, indicating that the orthorhombic perovskite phase remains the dominant crystalline phase after Au-Pd modification (Figure 2a). No distinct reflections attributable to crystalline Au- or Pd-containing phases or other impurity phases are observed, possibly because of their limited contents, small sizes, or high dispersion. The well-defined LFO reflections also indicate good crystallinity. EDS elemental mapping further confirms the presence of Au, Pd, La, Fe, and O (Figure 2b–f). The Au and Pd signals are relatively uniformly distributed throughout the examined region, without obvious large-scale enrichment. Together, these results support the successful modification of LFO with Au and Pd without altering its dominant crystalline phase.

Figure 2. (a) X-ray diffraction pattern of Au-Pd/LFO; (b–f) EDS elemental mapping images of the corresponding region.

The SEM images in Figure 3a–d show the morphology of Au-Pd/LFO at different magnifications. At low magnification, the material consists of irregular micrometer-scale agglomerates with rough surfaces. Higher-magnification images reveal that these agglomerates are composed of closely packed, quasi-spherical nanograins with apparent sizes mainly in the range of 50–70 nm. Abundant voids are also observed between adjacent grains, providing pathways for gas molecules to penetrate the material and access the reaction surface. The N₂ adsorption–desorption isotherm of Au-Pd/LFO exhibits a pronounced hysteresis loop, indicating its mesoporous characteristics (Figure 3e). The BJH pore-size distribution further suggests the presence of larger intergranular pores in addition to mesopores, consistent with the open particulate structure observed by SEM.

Figure 3. (a–d) SEM images of Au-Pd/LFO at different magnifications; (e) N₂ adsorption–desorption isotherms and pore size distributions of Au-Pd/LFO; (f) Specific surface area characterization diagram of four samples.

Figure 3f compares the BET specific surface areas of the four materials. The specific surface area of LFO is below 5 m² g⁻¹, whereas those of Au/LFO and Pd/LFO reach 8.53 and 7.62 m² g⁻¹, respectively. Au-Pd/LFO exhibits the largest specific surface area of 13.26 m² g⁻¹, approximately 1.55 and 1.74 times those of Au/LFO and Pd/LFO, respectively. Au-Pd/LFO therefore provides the largest gas-accessible surface area among the four materials. The combination of a larger specific surface area and abundant intergranular voids facilitates the access of oxygen and n-butanol molecules to the material surface and their subsequent adsorption and reaction, providing a structural basis for the enhanced sensing response.

The XPS spectra in Figure 4 provide detailed insights into the chemical states and surface compositions of the prepared samples. Figure 4a shows the Au 4f spin–orbit doublet, with the Au 4f_7/2 and Au 4f_5/2 peaks centered at approximately 84.7 and 88.3 eV, respectively, confirming that gold exists mainly as metallic and weakly oxidized nanoclusters on the surface. Figure 4b displays mixed Pd⁰ (335.3, 341.2 eV) and Pd²⁺ (336.8, 343.4 eV) signals, indicating partial electron transfer from Pd to the lattice-an electronic configuration known to promote low-temperature catalytic activity. The dominant Au⁰ and Pd⁰ components indicate that both noble metals remain largely in their metallic state after calcination. Metallic Au acts as an electron sink, forming Schottky contacts with LaFeO₃ and accelerating charge transfer, while metallic Pd supplies fast adsorption–desorption sites for oxygen and promotes low-temperature activation of alcohol molecules. Their coexistence therefore builds a dual catalytic/electronic network that favors rapid redox cycling on the material’s surface. Figure 4c–f compare the O 1s spectra of pristine LFO, Au/LFO, Pd/LFO and Au-Pd/LFO, respectively. Three deconvoluted components are evident: lattice oxygen (O_L), defect-related oxygen (O_V) and chemisorbed oxygen (O_C). Compared with pristine LFO, Au/LFO and Pd/LFO show more pronounced contributions from the O_V and O_C components relative to O_L. This trend is more evident in Au-Pd/LFO, suggesting changes in the surface oxygen environment following noble-metal modification. These changes may influence oxygen adsorption and oxygen-mediated surface reactions, thereby contributing to the enhanced n-butanol sensing response.

Figure 4. High-resolution XPS spectra of the prepared samples. (a) Au 4f spectrum of Au-Pd/LFO; (b) Pd 3d spectrum of Au-Pd/LFO; (c–f) O 1s spectra of LFO, Au/LFO, Pd/LFO and Au-Pd/LFO.

3.2. Gas-Sensing Properties

The operating temperature strongly affects gas adsorption, surface reaction, and desorption, resulting in volcano-shaped response–temperature profiles for the four materials (Figure 5a). LFO, Au/LFO, and Pd/LFO reach their maximum responses of approximately 2.87, 10.53, and 9.26 at 200 °C, respectively. In contrast, Au-Pd/LFO reaches a maximum response of 16.92 at a lower temperature of 180 °C, corresponding to approximately 5.90, 1.61, and 1.83 times the respective maximum responses of LFO, Au/LFO, and Pd/LFO. Thus, Au-Pd co-modification not only markedly enhances the response toward n-butanol but also lowers the optimal operating temperature from 200 °C to 180 °C. At relatively low temperatures, insufficient activation of adsorbed oxygen species and n-butanol limits the surface reaction. At temperatures above the optimum, accelerated desorption reduces the effective surface coverage of adsorbed oxygen and n-butanol, leading to a decrease in response.

Figure 5. Gas-sensing performance of LFO, Au/LFO, Pd/LFO, and Au-Pd/LFO. (a) responses toward 1 ppm n-butanol at different operating temperatures; (b) baseline resistances at different operating temperatures; (c) responses toward 0.1–10 ppm n-butanol; (d) dynamic responses of Au-Pd/LFO toward different n-butanol concentrations; (e) response and recovery times toward 1 ppm n-butanol; (f) responses toward 1 ppm n-butanol under different relative humidities; (g) long-term response stability over 30 days; and (h) responses toward n-butanol and various interfering gases at 1 ppm. Data in panels (c,e–g) are presented as the mean ± standard deviation of three repeated measurements under each test condition.

The baseline resistances of all four materials decrease with increasing operating temperature (Figure 5b). Au-Pd/LFO exhibits a relatively low baseline resistance at lower temperatures, while the resistance values of the four materials gradually become closer near 180 °C. Above 200 °C, the resistance of Au-Pd/LFO changes only slightly, whereas those of LFO, Au/LFO, and Pd/LFO continue to decrease. Overall, Au-Pd/LFO exhibits a weaker dependence of baseline resistance on operating temperature, particularly in the high-temperature region.

The responses of all four materials increase monotonically as the n-butanol concentration increases from 0.1 to 10 ppm (Figure 5c). At 10 ppm, LFO, Au/LFO, Pd/LFO, and Au-Pd/LFO exhibit responses of approximately 14.6, 51.2, 46.4, and 68.1, respectively. Throughout the tested concentration range, Au-Pd/LFO consistently exhibits the highest response. In the low-concentration region, the relatively large response increment may be associated with the numerous unoccupied surface-active sites available for n-butanol adsorption and surface reaction. As the n-butanol concentration further increases, these sites are progressively occupied, reducing the number of available sites and gradually decreasing the response increment. Notably, Au-Pd/LFO exhibits a clearly distinguishable response at the lowest tested n-butanol concentration of 0.1 ppm.

Table 1 summarizes the n-butanol sensing performance of representative metal-oxide materials, with reported gas concentrations ranging from 1 to 100 ppm and operating temperatures from 130 to 262 °C. Au-Pd/LFO exhibits a response of 16.92 to 1 ppm n-butanol at 180 °C and produces a distinguishable response at the lowest tested concentration of 0.1 ppm, demonstrating its capability for low-concentration n-butanol sensing.

Table 1. Comparison of the n-butanol sensing performance of Au-Pd/LFO with representative metal-oxide sensing materials reported in the literature.

Material	Temperature (°C)	Concentration (ppm)	Response	Ref.
4 wt% Au/LFO	225	100	115	[17]
Au–Pd/WO3	200	50	91	[25]
Co3O4/LFO-1	150	1	6.54	[36]
1 wt% Pt/LFO	175	100	~400	[37]
NiO/Fe2O3	200	10	4.2	[38]
Fe2O3-loaded NiFe2O4	130	10	29.7	[39]
ZnO/In2O3-2	262	100	302	[40]
Au–Pd/LFO	180	1	16.92	Present

Note: $\mathrm{Response}={\displaystyle \raisebox{1ex}{$R_g$}\!\left/ \!\raisebox{-1ex}{$R_a$}\right.}\mathit{or}{\displaystyle \raisebox{1ex}{$R_a$}\!\left/ \!\raisebox{-1ex}{$R_g$}\right.}$.

The stepped-concentration response curve further demonstrates the dynamic sensing behavior of Au-Pd/LFO toward n-butanol (Figure 5d). During each exposure, the response increases and gradually approaches a stable level, with the response level increasing as the n-butanol concentration rises. After n-butanol is removed and air is reintroduced, the response returns close to its initial baseline. These consistent response–recovery transitions demonstrate the good reversibility and repeatable sensing behavior of Au-Pd/LFO over successive exposure–purge cycles.

Figure 5e compares the response and recovery times of the four materials toward 1 ppm n-butanol. LFO exhibits the longest response and recovery times, whereas Au/LFO and Pd/LFO show markedly faster response–recovery processes. The response and recovery times of Au-Pd/LFO lie between those of pristine LFO and the two single-metal-modified materials. This behavior may be associated with more extensive n-butanol adsorption and surface reaction, which require a relatively longer desorption and surface re-equilibration process after n-butanol removal. Overall, Au-Pd/LFO achieves a favorable balance between response magnitude and response–recovery characteristics.

The responses of all four materials decrease as the RH increases from 20% to 90%, although the extent of the response loss differs markedly among the materials (Figure 5f). At 50% RH, LFO retains approximately 74.2% of its response measured at 20% RH, whereas Au/LFO and Pd/LFO retain approximately 90.8% and 87.8%, respectively. In comparison, Au-Pd/LFO retains approximately 97.7% of its initial response at 50% RH and 93.2% at 60% RH. A more pronounced response decrease is observed for all four materials when the RH reaches 70%, followed by a sharp decline as RH increases from 70% to 80%. These results demonstrate that Au-Pd co-modification substantially improves humidity tolerance, enabling Au-Pd/LFO to maintain a high response under moderate-humidity conditions and broadening its applicability for practical n-butanol detection under variable-humidity conditions.

Figure 5g evaluates long-term stability by measuring the response toward 1 ppm n-butanol every other day over a 30-day period. LFO exhibits the most pronounced response fluctuations, with the variation exceeding 10%, indicating relatively poor long-term response stability. In comparison, the response variations of Au/LFO and Pd/LFO remain below 2%. Au-Pd/LFO displays an almost unchanged response throughout the test period, with a response variation below 1%, demonstrating the best long-term stability among the four materials. These results indicate that noble-metal modification markedly improves the long-term response stability of LFO, with Au-Pd co-modification providing the greatest improvement.

Figure 5h compares the responses of the four materials toward n-butanol and several interfering gases. All four materials exhibit higher responses toward n-butanol than toward the other tested gases, indicating a preferential response toward the target gas. Among the tested interferents, formaldehyde (HCHO) induces the most pronounced cross-response. Although its response remains lower than that toward n-butanol, the appreciable response induced by HCHO could substantially compromise the reliable identification of n-butanol when the two gases coexist.

These results indicate that the response magnitude of Au-Pd/LFO at a single operating temperature may be insufficient for the unambiguous identification of n-butanol in the presence of HCHO. Therefore, HCHO was selected as a representative interfering gas, and binary n-butanol/HCHO mixtures with different total concentrations and mixing ratios were prepared. The responses of Au-Pd/LFO were recorded at 160 °C, 180 °C, and 200 °C to exploit the different temperature dependences of the two gases and construct a multi-temperature response fingerprint. The resulting responses were subsequently transformed into temperature-normalized features designed to reduce the influence of total concentration and then used with a machine-learning regression model to predict the n-butanol fraction in the binary mixtures.

3.3. Machine-Learning-Assisted Analysis of Binary Mixtures

Binary n-butanol/HCHO mixtures with total concentrations of 0.5, 1.0, and 2.0 ppm were prepared at n-butanol ratios of 100:0, 75:25, 50:50, 25:75, and 0:100. The responses of Au–Pd/LFO were recorded at 160 °C, 180 °C, and 200 °C to extract composition-related information from the different temperature dependences of the two gases. To reduce the influence of total concentration on the absolute response magnitude, the response increment at 180 °C was used as the reference, and a temperature-normalized response feature was defined as:

F_T/180 = (S_T − 1)/(S₁₈₀ − 1)

where S_T denotes the response measured at temperature T, and T is 160, 180, or 200 °C. By definition, F_180/180 is fixed at 1.

Figure 6a shows the normalized temperature responses obtained at different n-butanol/HCHO ratios. As the n-butanol ratio changed from 0:100 to 100:0, F_160/180 decreased from 1.125 to 0.760, corresponding to a variation of approximately 0.365. In comparison, F_200/180 decreased from 0.946 to 0.858, with a variation of approximately 0.088. The composition-dependent variation of F_160/180 was therefore approximately 4.1 times that of F_200/180, providing a more pronounced distinction among the mixture compositions. The maximum standard deviation among the three total concentrations was only 0.026, indicating that the normalized temperature response was only weakly affected by the total concentration. The temperature-response differences among the mixture compositions may be associated with the temperature-dependent adsorption, activation, and surface oxidation of n-butanol and HCHO. F_160/180 was therefore used as the input feature for the subsequent composition analysis.

Figure 6. Composition analysis of binary n-butanol/HCHO mixtures using the multi-temperature response features of Au-Pd/LFO. (a) Normalized excess responses of mixtures with different n-butanol ratios at different operating temperatures; (b) relationship between F_160/180 and the n-butanol fraction under different total concentrations; (c) RBF-SVR-predicted versus actual n-butanol fractions; (d) comparison of the MAE values obtained using different input features in concentration-transfer tests.

At total concentrations of 0.5, 1.0, and 2.0 ppm, F_160/180 decreased monotonically with increasing n-butanol fraction (Figure 6b). Although the highest total concentration was four times the lowest, the three curves retained similar fraction–feature relationships. This consistent monotonic behavior allowed the same feature mapping to be applied to mixtures with different total concentrations.

The relationship between F_160/180 and the n-butanol fraction deviated from a strictly linear trend. A support vector regression model with a radial basis function kernel (RBF-SVR) was therefore employed. Its prediction function is expressed as:

$$\widehat{y}\left(x\right)={\sum }_{i=1}^{N_s}\left(a_i-a_i^{*}\right)K\left(x_i,x\right)+b$$

The RBF kernel is given by:

$$K\left(x_i,x\right)=\exp \left[-\gamma {|x_i-x|}^2\right]$$

where x is the standardized F_160/180 input, x_i denotes the corresponding support vectors, $\widehat{y}$ is the predicted n-butanol fraction, N_s is the number of support vectors, a_i and $a_i^{*}$ are the model coefficients, b is the bias term, and γ is the kernel parameter. After training with the 1.0 ppm data, the model was applied directly to predict the n-butanol fractions at 0.5 and 2.0 ppm without refitting or parameter adjustment.

As shown in Figure 6c, the predicted values increased with the actual n-butanol fraction at both total concentrations and were distributed close to the identity line. At 0.5 ppm, the mean absolute error (MAE) and root-mean-square error (RMSE) were 0.075 and 0.099, respectively, and R² reached 0.921. At 2.0 ppm, the MAE and RMSE decreased to 0.033 and 0.049, respectively, with an R² of 0.981. The fraction–feature relationship established at 1.0 ppm remained effective at 0.5 and 2.0 ppm, demonstrating the advantage of F_160/180 for composition analysis across different total concentrations.

Figure 6d compares the predictive performance of different input features using the same data split and linear-regression model. When the raw response at 160, 180, or 200 °C was used individually, the MAEs for the 0.5 and 2.0 ppm datasets remained approximately 0.49–0.50. Combining the raw responses at all three temperatures did not appreciably reduce the prediction error. When F_160/180 was used instead, the MAEs decreased to 0.065 and 0.063 at 0.5 and 2.0 ppm, respectively. Relative to the combination of the three raw responses, these values represent reductions of 86.9% and 87.3%, respectively. The MAEs obtained using F_200/180 were 0.230 and 0.162, which were higher than those obtained using F_160/180, consistent with its smaller composition-dependent variation in Figure 6a. Combining F_160/180 and F_200/180 produced MAEs of 0.064 and 0.060, respectively, which were close to those obtained using F_160/180 alone. Thus, F_160/180 retained most of the composition-related information. The comparison shows that the improvement in concentration-transfer prediction arose primarily from the temperature-normalized feature rather than from simply increasing the number of raw responses or input features.

3.4. Gas-Sensing Mechanism

The proposed gas-sensing mechanism of Au-Pd/LFO is schematically illustrated in Figure 7. The resistance variation of Au-Pd/LFO toward n-butanol can be interpreted using the surface-adsorbed oxygen model commonly applied to p-type LFO [11,14,15]. In air, oxygen molecules adsorb on the material surface and capture electrons from LFO, forming ionized adsorbed oxygen species. The corresponding processes can be represented by the following simplified reactions:

O₂(g) → O₂(ads)

$$O_2\left(\mathit{ads}\right)+e^{-}\to O_2^{-}\left(\mathit{ads}\right)$$

$$O_2^{-}\left(\mathit{ads}\right)+e^{-}\to 2O^{-}\left(\mathit{ads}\right)$$

Figure 7. Schematic illustration of the gas-sensing mechanism of Au-Pd/LFO toward n-butanol.

At 180 °C, O⁻ is commonly considered one of the major reactive oxygen species within the conventional ionosorption model [41]. Electron capture by the adsorbed oxygen species increases the near-surface hole concentration and leads to the formation of a hole accumulation layer. Consequently, the near-surface conductivity increases, resulting in a lower resistance in air than during exposure to n-butanol.

When n-butanol molecules reach the surface of Au-Pd/LFO, they adsorb at available surface sites and subsequently react with the ionized oxygen species. The idealized overall reaction for the complete oxidation of n-butanol can be expressed as [15,17]:

C₄H₉OH + 12O⁻(ads) → 4CO₂ + 5H₂O + 12e⁻

The electrons released by this reaction enter the sensing material and recombine with holes, thereby decreasing the near-surface hole concentration and weakening the hole accumulation layer. Consequently, the conductivity decreases, and the resistance increases, giving a response greater than 1. After n-butanol is removed, oxygen molecules are re-adsorbed on the material surface and capture electrons. The near-surface hole concentration is gradually restored, and the resistance returns to a value close to its initial baseline. This process is consistent with the reversible response–recovery behavior observed in Figure 5d.

The larger accessible surface area and intergranular voids of Au-Pd/LFO may facilitate gas adsorption and transport, while its surface oxygen-related sites may contribute to oxygen-mediated surface reactions. Pd-containing species may promote the adsorption and activation of oxygen and n-butanol and facilitate the spillover of activated species across the LFO surface [19,24]. Au-related interfaces may modulate the local carrier distribution and the electrical response to these reactions [17,20,21,22]. Together, these structural, catalytic, and interfacial effects enhance the n-butanol response and lower the optimal operating temperature.

HCHO can also induce resistance changes through reactions with surface-adsorbed oxygen, accounting for the cross-response observed in Figure 5h. The different temperature dependences of the n-butanol and HCHO responses may reflect differences in their adsorption, activation, and surface oxidation processes. This contrast provides the basis for the composition-sensitive temperature-normalized feature used in Section 3.3.

4. Conclusions

To address the limited n-butanol response of LFO and the difficulty in resolving mixture composition caused by the cross-response to HCHO, this study enhanced the sensing performance of LFO through Au-Pd co-modification and extracted composition information from binary n-butanol/HCHO mixtures using multi-temperature response analysis. Au-Pd/LFO exhibited a response of 16.92 to 1 ppm n-butanol at 180 °C, approximately 5.90 times the maximum response of pristine LFO, while lowering the optimal operating temperature from 200 °C to 180 °C.

The larger accessible gas–solid interfacial area of Au-Pd/LFO and its higher relative fractions of defect-related and adsorbed-oxygen-related components may provide favorable conditions for the adsorption of oxygen and n-butanol and the subsequent surface reactions. The catalytic and interfacial effects introduced by Au and Pd further strengthen the resistance modulation associated with these reactions. To resolve the cross-response to HCHO, the temperature-normalized feature F_160/180 was constructed from the distinct temperature-response behaviors of n-butanol and HCHO as a descriptor of mixture composition. The RBF-SVR model developed using the 1.0 ppm data yielded MAEs of 0.075 and 0.033 at total concentrations of 0.5 and 2.0 ppm, respectively, for the five predefined n-butanol fractions. This work offers a viable strategy for mixed-gas component analysis using a single metal-oxide sensor without resorting to multi-sensor arrays.