Smart IoT-Enabled Embedded Platform for Real-Time Energy Efficiency Assessment in Heat Pumps ^†

Abstract

This study presents the design, implementation, and field validation of a low-cost Internet of Things (IoT)-enabled embedded platform for real-time energy efficiency assessment of heat pump systems. The platform integrates an ESP32-based microcontroller with calibrated temperature, flow rate, and electrical current sensors to automate the determination of the coefficient of performance (COP). Unlike conventional studies that rely on high-cost instrumentation or laboratory-only trials, this work emphasizes a portable and scalable solution deployable in real environments. A three-month continuous deployment demonstrated high measurement accuracy against certified reference instruments, achieving mean absolute error (MAE) values of 0.29 °C (temperature), 0.0027 L/min (flow rate), and 0.110 A (current). Under steady-state conditions, the daily COP ranged from 1.43 to 4.43, diverging from the nominal manufacturer value of 3.83 and confirming the influence of operational and environmental factors reported by field studies. These results indicate that the platform serves not only as a diagnostic and monitoring tool but also as a replicable framework for predictive maintenance, operational optimization, and sustainable energy management in both grid-connected and off-grid applications, bridging the gap between engineering implementation and scientific assessment of heat pump performance.

1. Introduction

Strengthening energy management in thermal systems improves efficiency through systematic optimization and informed operation [1]. This requires accessible, accurate monitoring to characterize conditions, quantify losses, and support data-driven decisions [2], with continuous observation of key variables limiting degradation [3]. A widely used indicator is the coefficient of performance (COP), relating delivered thermal energy to electrical input and enabling comparative assessment across technologies and regimes [4].

Heat pumps deliver efficient climate control and low-temperature heating [5] with relatively low electricity use [6], yet in-use performance depends on ambient temperature, flow rate, and electrical load [7]. Field studies report persistent gaps between nameplate and in situ performance due to environmental variability, cycling, and non-ideal control [4,5,6]. High-cost instrumentation, short campaigns, and limited redundancy further constrain replicability in resource-constrained contexts [3,8], motivating a low-cost, validated framework that couples calibrated sensing with on-device filtering, preserves data via redundant local/cloud paths, and enables long-term in situ deployment using widely available components [9]. Three strands of related work frame this need: (i) in situ performance and data-driven assessments that document COP gaps and motivate continuous sensing [3,4,5,6,7,8]; (ii) edge/embedded acquisition with ESP32-class MCUs and calibrated low-cost sensors enabling persistent field sensing with on-device preprocessing and time-stamped local logging [9,10,11,12,13,14,15,16,17]; and (iii) cloud backends that provide dashboards, alarms, analytics (including automated COP), and data integrity [18,19]. Within this landscape, the ESP32-based platform targets cost, calibrated accuracy, and deployment flexibility. Versus 8-bit MCUs lacking integrated Wi-Fi and compute headroom, the ESP32 enables synchronized sampling, light on-device filtering, and reliable telemetry for persistent field sensing [9]; compared with single-board computers, it offers similar functionality with lower power/maintenance and bill of materials while supporting autonomous microSD + RTC logging [14,15,16]; and relative to proprietary data acquisition (DAQs), a DS18B20, YF-S201, and SCT-013-030 chain preserves simple installation, field-grade accuracy, and cloud interoperability [18,19]. In deployment, the system achieved sub-degree temperature error, ≤0.01 $\mathrm{L}{\min }^{-1}$ flow error, and ∼0.1 $\mathrm{A}$ current error with redundant local/cloud handling.

This study addresses four technical challenges: achieving field-grade accuracy with low-cost temperature/flow/current sensing under real operating conditions [10,11,12,13]; ensuring time alignment and lossless long-term records under intermittent connectivity via RTC time-stamping, local logging, and cloud buffering/retry [9,14,15,16]; performing regime-aware analysis that separates steady versus transient behavior and quantifies factor influence on COP [20,21]; and adopting a resource-aware design (sampling cadence, power, and bill of materials) to enable minimally intrusive, replicable retrofits in constrained settings [8,9]. The main contributions of this work are as follows:

• A low-cost, portable acquisition system integrating calibrated temperature (DS18B20, Maxim Integrated, San José, CA, USA), flow-rate (YF-S201, Foshan Shunde Zhongjiang Energy Saving Electronics, Foshan, China), and current (SCT-013-030, Beijing YaoHuaDechang Electronic Co., Ltd., Beijing, China) sensing with on-device filtering, delivering sub-degree/≤0.01 $\mathrm{L}{\min }^{-1}$/∼0.1 $\mathrm{A}$ accuracy, and flexible deployment (autonomous microSD + RTC, Wi-Fi telemetry).
• A dual data-management strategy—local microSD logging with reliable time-stamping and real-time cloud transmission—that provides redundancy, auditability, and remote diagnostics.
• A three-month in situ deployment on a working heat pump, reporting accuracy against certified references and analyzing daily COP under steady-state and transient conditions, thereby linking engineering implementation with scientific assessment.

2. Methodology

This section details the research design followed to develop, implement, and validate a low-cost, IoT-enabled embedded platform for real-time energy-efficiency assessment of heat pump systems. The methodology is structured to ensure (i) clear hardware–software integration for in situ sensing, (ii) traceable data handling and redundancy for auditability, and (iii) calibration and validation against certified references to quantify accuracy and robustness under real operating conditions.

2.1. System Overview

The platform is built on an ESP32 (dual core, integrated Wi-Fi, and low power), with a sensing chain targeting COP drivers: DS18B20 temperature probes (−55–125 °C), a YF-S201 Hall flow sensor (1–$30\mathrm{L}{\min }^{-1}$), and an SCT-013-030 non-invasive current transformer (to $30\mathrm{A}$). Sensors are mounted at the water-loop inlet/outlet (temperature), on the primary line (flow), and on the compressor feed (current), yielding a retrofit-friendly, minimally intrusive layout. The ESP32 performs synchronous sampling with light on-device filtering prior to persistence and uplink.

Data handling follows a dual path: local microSD logging for auditability [22] with RTC time-stamping via a DS3231 [23], and cloud transmission over Wi-Fi (dashboards, alarms, and automated COP). This preserves continuous acquisition through temporary outages and enables backfilling on reconnection. The complete setup—heat pump, sensors, acquisition unit, and monitoring devices—is shown in Figure 1.

2.2. Measurement and Data Flow

The platform runs a continuous acquisition cycle with a fixed sampling period $T_s=60$ s, chosen as a trade-off between dynamic fidelity and resource use: with hydronic/thermal time constants of ∼3–10 min and compressor cycling periods ≥ 5 min, this cadence affords ≥3× oversampling of the dominant dynamics while capping data volume (1440 records/day, ≈0.1–0.2 MB/day CSV) and wireless duty (uplinks are batched), enabling long, reliable field deployments. Each record is time-stamped by the RTC to guarantee temporal alignment across sensors and power cycles. Lightweight on-device smoothing is applied before persistence and uplink to attenuate spikes while preserving step changes. Data handling follows a dual-path strategy: (i) local logging—each time-aligned record (values, timestamp, and status flags) is appended to a CSV file on the microSD card to ensure auditability and backfilling during outages—and (ii) cloud transmission—records are queued and sent through the ESP32’s Wi-Fi interface to Ubidots for dashboards, alarms, and automated COP computation; unsent entries remain buffered and are retried until acknowledgment. Range checks and missing-data guards are executed on-device prior to commit, and housekeeping metadata (e.g., uptime counters) are appended periodically for traceability. The acquisition loop avoids blocking operations; retransmissions are decoupled from sampling to minimize jitter and preserve timing determinism. Figure 2 summarizes the measurement and data flow.

2.3. Calibration and Experimental Setup

Calibration and validation targeted accuracy, repeatability, and robustness. A priori thresholds: temperature MAE ≤ 0.3 °C, flow MAE $\le 0.01\mathrm{L}{\min }^{-1}$, current MAE $\le 0.15\mathrm{A}$. References—NOVUS + Type K bath (−50–130 °C), Omega FLR-1000 (1–$30\mathrm{L}{\min }^{-1}$,Omega Engineering, Norwalk, USA), Fluke analyzer (to $30\mathrm{A}$). Static calibration used ∼25/45/75 °C for temperature and 6–8 points for flow/current; at each setpoint $n=10$ stabilized replicates were acquired; a linear or low-order polynomial was fitted per channel and coefficients were stored in firmware, while MAE/MAPE on held-out points; and repeatability was ensured from replicate SD.

Dynamic validation exercised flow steps, compressor start/stop, and temperature ramps without retuning. Sensors were installed at the hydraulic inlet/outlet (temperature), the primary line with straight runs (flow), and the compressor feed (current). The ESP32 sampled synchronously at $T_s=60$ s; each record was median-filtered, five-point averaged, RTC time-stamped, stored locally (CSV), and queued for cloud upload with retry; a lightweight quality gate (range, missing-data, and status flags) preceded commit. Firmware fault monitoring runs continuously: sensor sanity/stasis (range checks and stuck-value), process thresholds (Q, $\left|\mathsf{\Delta }T\right|$, $T_{\mathrm{out}}$), electrical anomalies ($I_{\mathrm{rms}}$, excessive cycling), sustained low-COP, and connectivity/power supervision; on violation, a status flag is latched, a cloud alarm is raised, local logging continues with retries, and a watchdog preserves timing.

Ambient temperature and relative humidity were logged with a thermo-hygrometer (accuracy ±0.5 °C, $\pm 2\%$RH) positioned ∼0.5 m from the outdoor intake and shaded, at the same cadence ($T_s=60$ s); over the three-month campaign, ambient ranged from 12 to 28 °C (median 19 °C) and RH 45–$88\%$ (median $63\%$), providing the operating context for Section 3.2.

3. Results and Discussion

3.1. Sensor Accuracy and System Stability

After calibration, all sensing channels met the a priori acceptance thresholds defined in Section 2.3. Absolute and relative accuracies were quantified using Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE), respectively [24,25]. For temperature, DS18B20 probes evaluated at approximately 25, 45, and 75 °C showed MAE values of 0.29 °C, 0.28 °C, and 0.09 °C, with MAPE below $1.2\%$ at all points, confirming suitability for COP estimation under field conditions (Figure 3).

For flow, a comparison of the YF-S201 against the reference meter yielded MAE $=0.0027\mathrm{L}{\min }^{-1}$ and MAPE $=0.16\%$ across the 1–$30\mathrm{L}{\min }^{-1}$ operating range, indicating stable behavior over the full measurement window (Figure 4).

For electrical current, the SCT-013-030 exhibited MAE $=0.110\mathrm{A}$ and MAPE $=1.19\%$ over the tested band, showing close agreement with the reference analyzer (Figure 5).

Regarding system connectivity, the Wi-Fi link to the Ubidots backend achieved a success rate of $96\%$ with an average connection time of $3.8$ s; the autoreconnect and local buffering mechanisms maintained $0\%$ data loss during temporary outages, ensuring continuity of the time series (Figure 6).

3.2. COP Analysis and Energy Performance

A three-month campaign with one-minute sampling enabled computation of the instantaneous coefficient of performance (COP) under steady-state [20] and transient conditions [21]. COP was defined as delivered thermal power over electrical input; steady-state intervals were detected from quasi-constant flow and temperature plateaus, while transients covered start-up, load changes, and off-design operation. During the three-month campaign, no hard faults were triggered; brief connectivity dips were absorbed by the buffering/retry logic without data loss. For quantifying factor influence, we computed Spearman’s $\rho$ between daily COP and ambient temperature ($T_{\mathrm{amb}}$), inlet temperature ($T_{\mathrm{in}}$), flow rate (Q), compressor duty cycle (D), and relative humidity (RH) and fitted a multiple linear regression with standardized predictors to obtain coefficients ($\beta$) and $R^2$.

In steady-state, $T_{\mathrm{in}}$ and Q showed positive association with COP ($\rho =+0.58$ and $+0.41$, both $p<0.001$; $\beta =+0.44$ and $+0.28$), whereas $T_{\mathrm{amb}}$ and D were negative drivers ($\rho =-0.36$ and $-0.33$, $p\le 0.001$; $\beta =-0.23$ and $-0.19$); RH was weak ($\rho =-0.12$, $p=0.21$; $\beta =-0.06$). The model explained $R^2=0.52$ of variance.

In transients, D and $T_{\mathrm{amb}}$ dominated negatively ($\rho =-0.52$ and $-0.41$, $p<0.001$; $\beta =-0.49$ and $-0.31$), while $T_{\mathrm{in}}$ and Q were smaller positives ($\rho =+0.26$, $p=0.014$; $\beta =+0.18$ and $\beta =+0.12$); RH remained negligible ($\rho =-0.15$, $p=0.12$; $\beta =-0.05$), with $R^2=0.47$.

Under steady-state flow, the daily average COP ranged from $1.43$ to $4.43$ (mean $2.63$), consistent with the positive roles of $T_{\mathrm{in}}$ and Q and the negative impact of $T_{\mathrm{amb}}$ and D; periods with $\mathrm{COP}\approx 4.0$ indicate operation within the design envelope (Figure 7). During transients, COP dropped to $0.1$–$1.7$ (mean $1.11$), reflecting unstable flow and compressor cycling—aligned with the stronger negative influence of D and $T_{\mathrm{amb}}$ (Figure 8). Relative to the manufacturer’s nominal $\mathrm{COP}=3.83$, both regimes confirm that in situ performance seldom matches nameplate values, in line with prior field studies [26,27].

3.3. System Usability and Deployment Potential

The platform is compact, portable, and autonomous, enabling rapid installation in laboratory and field settings. Its modular design and integrated battery operation support deployments in remote or off-grid locations. During the three-month campaign, the system maintained continuous acquisition with complete data retention: local CSV logging and the buffering logic described in Section 2.2 ensured integrity during temporary outages, while the wireless link sustained high success rates with minimal operator intervention beyond initial setup.

Power at the 5 V input was measured in three states: idle (MCU on, no Wi-Fi TX) $\approx 90$ mA ($0.45$ W), acquisition (sampling + filtering) $\approx 110$ mA ($0.55$ W), and uplink (Wi-Fi TX) $\approx 180$ mA ($0.90$ W), with duty cycles of 70%, 25%, and 5%. This gives $E_{\mathrm{day}}\approx 12.0$ Wh/day ($\sim 0.50$ W average), so a 48 Wh battery lasts ∼4 days off-grid. The bill of materials totaled USD 49, or USD 16.3 per monitored channel (temperature, flow, current).

Table 1. Daily energy and cost in representative low-cost configurations.

Compute Core	BoM (USD)	Energy (Wh/day)	Notes
ESP32 (this work)	USD 49	12	Wi-Fi on-board; microSD + RTC
8-bit MCU + ESP8266 Wi-Fi	USD 60	10	External Wi-Fi; lower compute headroom
Single-board computer (SBC)	USD 75	36	Higher power; OS maintenance

compares low-cost configurations using the same sensing chain and functionality.

With identical sensors and functionality, the ESP32 delivers the lowest BoM, integrated telemetry, efficient filtering, and much lower energy use than SBC-based monitors, confirming—together with the calibrated accuracy in Section 3.1—its cost-effectiveness and suitability for scalable, long-term COP monitoring.

4. Conclusions

This study demonstrated a low-cost, IoT-enabled embedded platform—ESP32 with calibrated temperature, flow-rate, and current sensing—for real-time in situ energy-efficiency assessment of heat pumps. A three-month deployment validated measurement fidelity (MAE: 0.29 °C temperature, $0.0027\mathrm{L}{\min }^{-1}$ flow, $0.110\mathrm{A}$ current) and end-to-end robustness: the dual-path data architecture (local microSD + RTC plus cloud with buffering/retry) preserved complete datasets through intermittent connectivity, and firmware fault monitoring operated continuously without hard-fault triggers. Field results quantified performance and drivers. Steady-state daily $\mathrm{COP}$ ranged from $1.43$ to $4.43$ (mean $2.63$), while transients yielded $0.1$–$1.7$ (mean $1.11$), confirming gaps to the nominal $\mathrm{COP}=3.83$. A compact correlation/regression analysis showed $T_{\mathrm{in}}$ and Q as positive predictors and $T_{\mathrm{amb}}$ and duty cycle as negative drivers, supporting targeted diagnostics and operational optimization. The platform achieved these outcomes with a bill of materials of USD 49 and an energy budget of ∼12 Wh day⁻¹, outperforming like-for-like low-cost alternatives in price–power trade-offs while retaining integrated Wi-Fi and autonomous logging. The compact, modular design is retrofit-friendly and portable, enabling rapid deployment for research and industrial audits, continuous monitoring, and predictive maintenance. Limitations include a single site and one-minute sampling cadence. Future work will broaden instrumentation (e.g., pressure/heat-meter inputs), extend datasets across climates and operating envelopes, incorporate predictive analytics and supervisory control, and deepen interoperability with cloud services to scale in situ COP optimization.