A single accelerometer with limited performance often struggles to meet the increasingly complex demands of vibration measurement. Complementarity among various accelerometers shows promise for performance improvement through integration combined with fusion. Inspired by this, a method of an integrated array accelerometer (IAA) with an expanded dynamic range (DR) based on the adaptive data selection fusion (ADSF) technique is proposed in this article. Initially, this article presents the design of the architectural and operating principle of the IAA, along with a spatiotemporal calibration method to rectify misaligned data. Subsequently, aiming at the difficulties of achieving lossless and efficient fusion for multiple accelerometers with different DRs, a novel ADSF algorithm is developed, comprising: 1) a preprocessing process that incorporates the single-step Kalman filter to correct outliers and errors; 2) an adaptive affine-weighted fusion (AAWF) algorithm based on neural network and logarithmic mapping, complemented by a robust virtual label-based training method; and 3) a channel decision mechanism (CDM) utilizing a multilevel gradient prediction technique. The ADSF model was trained and deployed into the IAA prototype. Experimental results demonstrate that the prototype effectively operates within an expanded DR of 160.22 dB (expanded by 133.35 times) and performs not inferior to its subaccelerometers in terms of accuracy and stability. The output data rate (ODR) can achieve up to three-axis at 857 Hz and single-axis at 1461 Hz, respectively. These outcomes not only confirm the proposed method’s effectiveness but also show its potential applicability to various other sensor types.

中文翻译：

---

基于自适应数据选择融合的扩展动态范围集成阵列加速度计的设计与实现

性能有限的单个加速度计通常难以满足日益复杂的振动测量需求。各种加速度计之间的互补性显示出通过集成与融合来提高性能的希望。受此启发，本文提出了一种基于自适应数据选择融合（ADSF）技术的具有扩展动态范围（DR）的集成阵列加速度计（IAA）方法。首先，本文介绍了 IAA 的架构和操作原理的设计，以及纠正未对齐数据的时空校准方法。随后，针对不同DR的多个加速度计实现无损高效融合的困难，开发了一种新颖的ADSF算法，包括：1）结合单步卡尔曼滤波器来校正异常值和误差的预处理过程； 2）基于神经网络和对数映射的自适应仿射加权融合（AAWF）算法，辅以鲁棒的基于虚拟标签的训练方法； 3)利用多级梯度预测技术的信道决策机制(CDM)。 ADSF 模型经过训练并部署到 IAA 原型中。实验结果表明，原型机在160.22 dB的扩展DR（扩展了133.35倍）内有效运行，并且在精度和稳定性方面表现不逊色于其子加速度计。输出数据速率（ODR）可分别达到三轴（857 Hz）和单轴（1461 Hz）。这些结果不仅证实了所提出方法的有效性，而且还表明了其对各种其他传感器类型的潜在适用性。